A Model Context Protocol (MCP) server that brings Firecrawl, the context API for AI agents, to MCP-compatible clients: search, scrape, and interact with the live web and get back clean Markdown or structured data.

Big thanks to @vrknetha, @knacklabs for the initial implementation!

• Search the web, optionally with full page content for each result
• Search indexes for developer sources (GitHub issues, merged pull requests, READMEs, and docs), US government sources, and scientific papers (PubMed, bioRxiv, medRxiv, arXiv)
• Scrape any URL into clean Markdown
• Parse PDFs and documents
• Crawl a site for content from many pages, or map it to discover its URLs
• Interact with pages in a real browser: click, type, fill forms, and navigate dynamic pages
• Run Agent jobs for multi-source research that returns structured data
• Monitor pages and get notified when their content changes
• Check current and historical Firecrawl credit usage
• Automatic retries and rate limiting
• Cloud and self-hosted support
• SSE support

Play around with our MCP Server on MCP.so's playground or on Klavis AI.

• Use firecrawl_scrape when you have a known URL and want its content as markdown, then extract the requested fields yourself.
• Use firecrawl_map when you need to discover URLs on a site without fetching their content.
• Use firecrawl_crawl when you need content from many pages under a site; set limit, includePaths/excludePaths, or maxDiscoveryDepth to bound it.
• Use firecrawl_search when you're starting from a query rather than a URL and want ranked web results. Results carry excerpts; fetch full pages with firecrawl_scrape on the URLs you need. scrapeOptions fetches every result's page in the same call and bills each as a scrape, so keep it for the full text of a few results (the search-only endpoint never fetches content).
• Use firecrawl_interact when a page needs a click, type, or navigate action before you can read it: pass a url for a fresh page or a scrapeId to continue on one you already scraped.
• Use the firecrawl_monitor_* tools when the same page needs to be checked on a recurring schedule with diffs and change alerts, rather than fetched once.
• Use firecrawl_credit_usage to check credits left or what credits were spent on, by month or day, optionally broken down by endpoint or API key.
• Consider something else when you need to hold a browser session open across many of your own steps with your own retry and termination logic: each firecrawl_interact call runs one prompt or code turn to completion and returns control. The session can persist across calls via scrapeId and ends with firecrawl_interact_stop, but you cannot drive it interactively step-by-step from the client side within a single call.

This server lists 31 tools when the full profile registers with default settings (feedback tools included, not running in local-keyless mode). Setting FIRECRAWL_NO_SEARCH_FEEDBACK=1 and/or FIRECRAWL_NO_ENDPOINT_FEEDBACK=1 removes the corresponding feedback tools and reduces this count, as does local keyless startup. For clients with a tool-slot limit: the hosted keyless endpoint (https://mcp.firecrawl.dev/v2/mcp, no API key) exposes only 3 (firecrawl_scrape, firecrawl_search, firecrawl_parse), and the dedicated search-only endpoint (https://mcp.firecrawl.dev/v2/mcp-search) exposes a fixed set of 11 tools (search, developer, government, and research search, Alexandria catalogue lookup and execution, and people and company search and enrichment).

Connect to the remote hosted server with no setup:

https://mcp.firecrawl.dev/v2/mcp

On the keyless free tier, scrape, search, and parse work without an API key (rate-limited). Other tools such as crawl, map, and agent still need a key.

Prefer OAuth or an API key whenever the human can sign up. It unlocks the full tool set and higher limits.

For an interactive account connection, configure your MCP client to use this server URL. This is an MCP endpoint, not a browser page; use the client's account-connection flow and do not add a second Firecrawl server entry when reconnecting:

https://mcp.firecrawl.dev/v2/mcp-oauth

For an API-key connection (for example, an unattended integration), keep the server URL as:

https://mcp.firecrawl.dev/v2/mcp

Then configure the client's secure header or secret setting with:

Authorization: Bearer <FIRECRAWL_API_KEY>

Never put an API key in the server URL. Never put an API key in an agent chat. Configure it directly in the client or secret manager. See the hosted MCP setup guide and the agent onboarding guide for client-specific instructions.

An integration that connects over HTTP can name itself with an X-Firecrawl-Origin header, for example X-Firecrawl-Origin: my-agent. The server reports it to Firecrawl as the request origin mcp-my-agent@<server version>, in place of the client name or User-Agent it would otherwise report. Use letters, digits and dashes: each run of other characters becomes one dash, leading and trailing dashes are dropped, and the name is cut at 48 characters. Tool results are unaffected; only the origin sent to the Firecrawl API (and passed to an embedder's beforeKeylessRequest) changes.

A fixed-scope search surface is also hosted at:

https://mcp.firecrawl.dev/v2/mcp-search

It exposes a fixed set of eleven tools: firecrawl_search, firecrawl_developer_search, firecrawl_gov_search, the four firecrawl_research_* tools, the two Alexandria tools firecrawl_find_tools and firecrawl_scrape, and the enrich tools people_company_search and people_company_enrich. Its firecrawl_search fetches no page content.

env FIRECRAWL_API_KEY=fc-YOUR_API_KEY npx -y firecrawl-mcp

npm install -g firecrawl-mcp

Configuring Cursor 🖥️ Note: Requires Cursor version 0.45.6+ For the most up-to-date configuration instructions, please refer to the official Cursor documentation on configuring MCP servers: Cursor MCP Server Configuration Guide

To configure Firecrawl MCP in Cursor v0.48.6

1. Open Cursor Settings
2. Go to Features > MCP Servers
3. Click "+ Add new global MCP server"
4. Enter the following code:

   {
     "mcpServers": {
       "firecrawl-mcp": {
         "command": "npx",
         "args": ["-y", "firecrawl-mcp"],
         "env": {
           "FIRECRAWL_API_KEY": "YOUR-API-KEY"
         }
       }
     }
   }

To configure Firecrawl MCP in Cursor v0.45.6

1. Open Cursor Settings
2. Go to Features > MCP Servers
3. Click "+ Add New MCP Server"
4. Enter the following:
   • Name: "firecrawl-mcp" (or your preferred name)
   • Type: "command"
   • Command: env FIRECRAWL_API_KEY=your-api-key npx -y firecrawl-mcp

If you are using Windows and are running into issues, try cmd /c "set FIRECRAWL_API_KEY=your-api-key && npx -y firecrawl-mcp"

Replace your-api-key with your Firecrawl API key. If you don't have one yet, you can create an account and get it from https://www.firecrawl.dev/app/api-keys

After adding, refresh the MCP server list to see the new tools. The Composer Agent will automatically use Firecrawl MCP when appropriate, but you can explicitly request it by describing what you need from the web. Access the Composer via Command+L (Mac), select "Agent" next to the submit button, and enter your query.

Add this to your ./codeium/windsurf/model_config.json:

{
  "mcpServers": {
    "mcp-server-firecrawl": {
      "command": "npx",
      "args": ["-y", "firecrawl-mcp"],
      "env": {
        "FIRECRAWL_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

To run the server using Streamable HTTP locally instead of the default stdio transport:

env HTTP_STREAMABLE_SERVER=true FIRECRAWL_API_KEY=fc-YOUR_API_KEY npx -y firecrawl-mcp

Use the url: http://localhost:3000/mcp

In this mode the server only answers requests addressed to localhost, 127.0.0.1 or [::1] (on any port), plus the address in HOST when it isn't 0.0.0.0 or ::. A request whose Host header names another host, or whose Origin header names another site, gets 403, and CORS headers echo an allowed origin instead of *. GET and HEAD requests to /health are answered whatever the host, so health checks keep working, and the server logs a line naming FIRECRAWL_MCP_ALLOWED_HOSTS for every request it refuses. This stops web pages you visit from calling the local server through DNS rebinding or cross-origin requests and using its FIRECRAWL_API_KEY.

To reach the server under another name or address, such as a hostname behind a reverse proxy or a LAN IP, list it in FIRECRAWL_MCP_ALLOWED_HOSTS (comma-separated; ports are ignored). The same list decides which browser origins may call the server. A reverse proxy on the same machine can keep the default bind address:

env HTTP_STREAMABLE_SERVER=true FIRECRAWL_API_KEY=fc-YOUR_API_KEY \
  FIRECRAWL_MCP_ALLOWED_HOSTS=mcp.home.lan npx -y firecrawl-mcp

Clients on other machines also need the server to listen on a non-loopback address, so set HOST too. Every client that can reach the port is then served with the server's FIRECRAWL_API_KEY, so do this only on a network you trust:

env HTTP_STREAMABLE_SERVER=true FIRECRAWL_API_KEY=fc-YOUR_API_KEY HOST=0.0.0.0 \
  FIRECRAWL_MCP_ALLOWED_HOSTS=192.168.1.20 npx -y firecrawl-mcp

FIRECRAWL_MCP_ALLOWED_HOSTS=* turns the check off and restores the wildcard CORS header. Only do that when something in front of the server checks hosts and origins itself.

The ghcr.io/firecrawl/firecrawl-mcp-server image runs the same server as npx -y firecrawl-mcp. Every build from main is tagged latest, v<package version> and sha-<commit sha>. v<package version> moves to the newest build of that version, so pin sha-<commit sha> (or a digest) when you need a fixed image.

stdio (the default), for MCP clients that launch the server as a command:

docker run -i --rm -e FIRECRAWL_API_KEY=fc-YOUR_API_KEY ghcr.io/firecrawl/firecrawl-mcp-server:latest

{
  "mcpServers": {
    "firecrawl-mcp": {
      "command": "docker",
      "args": ["run", "-i", "--rm", "-e", "FIRECRAWL_API_KEY", "ghcr.io/firecrawl/firecrawl-mcp-server:latest"],
      "env": {
        "FIRECRAWL_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

Streamable HTTP, served at http://localhost:3000/mcp:

docker run --rm -p 127.0.0.1:3000:3000 \
  -e HTTP_STREAMABLE_SERVER=true \
  -e FIRECRAWL_API_KEY=fc-YOUR_API_KEY \
  ghcr.io/firecrawl/firecrawl-mcp-server:latest

The image sets HOST=0.0.0.0 and PORT=3000; override PORT (and the -p mapping) to listen elsewhere. The server uses the container's FIRECRAWL_API_KEY for requests that don't send their own credential, so the example publishes the port on 127.0.0.1 only; don't expose it more widely unless clients must supply their own key. Set FIRECRAWL_API_URL to use a self-hosted Firecrawl API. The container runs as the unprivileged node user. The server answers only localhost, 127.0.0.1 and [::1] in the Host and Origin headers, so the -p 127.0.0.1:3000:3000 mapping works as is; to reach it under another name, set -e FIRECRAWL_MCP_ALLOWED_HOSTS=<host> (see Streamable HTTP local mode).

For one-click installation, click one of the install buttons below...

For manual installation, add the following JSON block to your User Settings (JSON) file in VS Code. You can do this by pressing Ctrl + Shift + P and typing Preferences: Open User Settings (JSON).

{
  "mcp": {
    "inputs": [
      {
        "type": "promptString",
        "id": "apiKey",
        "description": "Firecrawl API Key",
        "password": true
      }
    ],
    "servers": {
      "firecrawl": {
        "command": "npx",
        "args": ["-y", "firecrawl-mcp"],
        "env": {
          "FIRECRAWL_API_KEY": "${input:apiKey}"
        }
      }
    }
  }
}

Optionally, you can add it to a file called .vscode/mcp.json in your workspace. This will allow you to share the configuration with others:

{
  "inputs": [
    {
      "type": "promptString",
      "id": "apiKey",
      "description": "Firecrawl API Key",
      "password": true
    }
  ],
  "servers": {
    "firecrawl": {
      "command": "npx",
      "args": ["-y", "firecrawl-mcp"],
      "env": {
        "FIRECRAWL_API_KEY": "${input:apiKey}"
      }
    }
  }
}

• FIRECRAWL_API_KEY: Your Firecrawl API key
  • Required when using cloud API (default)
  • Optional when using self-hosted instance with FIRECRAWL_API_URL
• FIRECRAWL_API_URL (Optional): Custom API endpoint for self-hosted instances
  • Example: https://firecrawl.your-domain.com
  • If not provided, the cloud API will be used (requires API key)
  • Over stdio, Alexandria tools are sent to this endpoint even when no API key is set here, and it answers for the account. Firecrawl Local signs them with the login it holds
• FIRECRAWL_INTEGRATION (Optional): Integration tag for apps that launch this server, such as an editor extension
  • Sent as the integration field of scrape, batch scrape, crawl, map, search, extract, agent, parse, interact and feedback requests
  • Use a value that starts with _, for example _zed

• FIRECRAWL_MCP_ALLOWED_HOSTS (Optional): Extra host names or addresses the server answers with HTTP_STREAMABLE_SERVER=true or SSE_LOCAL=true, comma-separated
  • localhost, 127.0.0.1, [::1] and a non-wildcard HOST are always allowed; ports are ignored
  • Requests whose Host or Origin header names any other host get 403, except GET and HEAD requests to /health
  • * turns the check off; not used with CLOUD_SERVICE=true, which answers every host

Firecrawl issues OAuth access tokens (fco_…) via the authorization server on firecrawl.dev. This MCP server forwards whichever credential it resolves to the Firecrawl API as Authorization: Bearer ….

• HTTP stream transports (CLOUD_SERVICE=true, HTTP_STREAMABLE_SERVER=true, or SSE_LOCAL=true): Clients should send Authorization: Bearer <fco_access_token> on MCP requests. An OAuth bearer token takes precedence over x-firecrawl-api-key / x-api-key when both are present.
• stdio: Use FIRECRAWL_OAUTH_TOKEN for a static access token, or keep using FIRECRAWL_API_KEY for an API key.

Use access tokens (fco_…) only. Refresh tokens (fcr_…) must be exchanged at the token endpoint, not passed to the scrape/search API.

CLOUD_SERVICE=true serves the tools over streamable HTTP on PORT (default 3000, bound to 0.0.0.0) with stricter defaults: every request must carry a credential (an API key or an OAuth access token), scrape is read-only (see the scrape tool below), and firecrawl_parse accepts uploads only.

FASTMCP_ENDPOINT selects the tool surface: /v2/mcp (default) for the full tool set, /v2/mcp-oauth for the account surface, or /v2/mcp-search for the search-only tools (this also requires FIRECRAWL_MCP_SEARCH_OAUTH_ONLY=true). With the full surface, a search-only instance also starts on port 3001; set FIRECRAWL_MCP_SEARCH_ENABLED=false to turn it off.

For cloud API usage:

export FIRECRAWL_API_KEY=your-api-key

For self-hosted instance:

# Required for self-hosted
export FIRECRAWL_API_URL=https://firecrawl.your-domain.com

# Optional authentication for self-hosted
export FIRECRAWL_API_KEY=your-api-key  # If your instance requires auth

Add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "mcp-server-firecrawl": {
      "command": "npx",
      "args": ["-y", "firecrawl-mcp"],
      "env": {
        "FIRECRAWL_API_KEY": "YOUR_API_KEY_HERE"
      }
    }
  }
}

Use this guide to select the right tool for your task:

• If you know the exact URL you want: use scrape, then extract the requested fields from its markdown
• If you have multiple known URLs: call scrape for each URL. If you specifically need one bulk API operation, use the Firecrawl API batch endpoint outside MCP.
• If you need to discover URLs on a site: use map
• If you want to search the web for info: use search
• If you have a programming question (a library, an API contract, an error message, a known bug): use developer search
• If you have a legal or regulatory question (a statute, regulation, code, court opinion, or other US government publication): use the Firecrawl Government Index
• If you need scientific papers (biomedical, life-science, clinical, or arXiv literature): use research tools, which search paper abstracts and full text. search with categories: ["research"] is a different thing: a website filter over ordinary web results.
• If you need multi-source research that returns structured data, do not know the URLs, or the answer spans several sites (an entity plus its fields, a list, a dataset): use agent
• If you want to analyze a whole site or section: use crawl (with limits!)
• If you need interactive browser automation (click, type, navigate): use interact with a URL for a fresh page, or scrape + interact when you already scraped the page or need tighter scrape control
• If you need data from a catalogued provider (Alexandria): search with sources: ["alexandria"], explore the library and read the contract with alexandria_library_search, and run it with alexandria_library_read

When using scrape, choose the right format:

• Markdown format: Read the page and extract the requested fields yourself.

Scrape content from a single URL with advanced options.

Best for:

• Single page content extraction, when you know exactly which page contains the information.

Not recommended for:

• Extracting content from multiple pages (use repeated scrape calls for known URLs, or map + scrape to discover URLs first, or crawl for full page content)
• When you're unsure which page contains the information (use search)

Common mistakes:

• Passing a list of URLs to one scrape call. Call scrape once per URL in MCP. If you specifically need one bulk API operation, use the Firecrawl API batch endpoint outside MCP.
• Requesting json or jsonOptions for new MCP scrape calls. These inputs are hidden from discovery but remain accepted for compatibility with cached tool definitions.

Choosing the right format:

• Markdown format: Read the returned content and extract the fields you need yourself.

Prompt Example:

"Get the product details from https://example.com/product."

Usage Example (read product details):

{
  "name": "firecrawl_scrape",
  "arguments": {
    "url": "https://example.com/product",
    "formats": ["markdown"]
  }
}

Usage Example (markdown format - when full content needed):

{
  "name": "firecrawl_scrape",
  "arguments": {
    "url": "https://example.com/article",
    "formats": ["markdown"],
    "onlyMainContent": true
  }
}

Usage Example (branding format - extract brand identity):

{
  "name": "firecrawl_scrape",
  "arguments": {
    "url": "https://example.com",
    "formats": ["branding"]
  }
}

Branding format: Extracts comprehensive brand identity (colors, fonts, typography, spacing, logo, UI components) for design analysis or style replication. Privacy: Set redactPII: true to return content with personally identifiable information redacted. With CLOUD_SERVICE=true: scrape is read-only. It takes no browser actions and cannot accept provider terms. A named profile loads saved browser state without saving changes to it, and the full endpoint's firecrawl_search treats scrapeOptions.profile the same way. To save browser state to a profile, open the page with firecrawl_interact (see below). An organization admin accepts terms in the dashboard.

Returns:

• Markdown, branding profile, or other formats as specified.

Map a website to discover all indexed URLs on the site.

Best for:

• Discovering URLs on a website before deciding what to scrape
• Finding specific sections of a website

Not recommended for:

• When you already know which specific URL you need (use scrape)
• When you need the content of the pages (use scrape after mapping)

Common mistakes:

• Using crawl to discover URLs instead of map

Prompt Example:

"List all URLs on example.com."

Usage Example:

{
  "name": "firecrawl_map",
  "arguments": {
    "url": "https://example.com"
  }
}

Returns:

• Array of URLs found on the site

Search the web and optionally extract content from search results.

Best for:

• Finding specific information across multiple websites, when you don't know which website has the information.
• When you need the most relevant content for a query

Not recommended for:

• When you already know which website to scrape (use scrape)
• When you need comprehensive coverage of a single website (use map or crawl)

Common mistakes:

• Using crawl or map for open-ended questions (use search instead)

Usage Example:

{
  "name": "firecrawl_search",
  "arguments": {
    "query": "remote work stipend policies at tech companies",
    "highlights": true,
    "limit": 5,
    "lang": "en",
    "country": "us"
  }
}

To get the full text of a few results in the same call, add "scrapeOptions": { "formats": ["markdown"], "onlyMainContent": true }. Each fetched result is billed as a scrape, so for most questions use the excerpts and firecrawl_scrape only the URLs you need.

Set highlights to true to request query-relevant highlights or false to keep the original search snippets. Omit it to use the API's default behavior.

Add "sources": ["alexandria"] for semantic tool discovery in data.tools, optionally mixed with web/news/images. A query is required. Use alexandria_library_search to explore the whole library and read contracts (firecrawl_find_tools on the search-only endpoint); see Alexandria Tools.

For scientific papers, see Research Tools: they search paper abstracts and full text, while categories: ["research"] here filters ordinary web results to research-affiliated websites.

Returns:

• Search results grouped by source (data.web, data.images, data.news), with optional scraped content, plus an id field. Every result carries a source and a 1-indexed position, numbered from 1 within its own group — copy the pair straight into firecrawl_search_feedback's valuableResults rather than counting array elements. Pass that id to firecrawl_search_feedback after you've used the results to refund 1 credit (search costs 2) and improve search quality.

Prompt Example:

"Compare remote work stipend policies across tech companies."

Sends structured feedback on a previous firecrawl_search result. The first feedback per search id refunds 1 credit and improves Firecrawl's search quality. Idempotent per search id.

Call this after every search you actually use (or that didn't help). Bad/partial feedback with missingContent is just as valuable as good feedback.

Opt out: set FIRECRAWL_NO_SEARCH_FEEDBACK=1 (or FIRECRAWL_DISABLE_SEARCH_FEEDBACK=1) in the environment when starting the MCP server. The firecrawl_search_feedback tool will not be registered, so agents can't call it. Team admins can also disable feedback server-side; in that case the tool is registered but always returns feedbackErrorCode: "TEAM_OPTED_OUT".

Most important field: missingContent. It's an array of specific pieces of content the agent expected to find but did not. One entry per missing topic; these aggregate across teams and tell us what to index next. For useful results, use valuableResults with the source/position pair stamped on each result — results are grouped and each group is numbered from 1 independently, so both are required. List every useful result (unlisted results are treated as not useful), and reserve valuableSources for useful URLs that were not among the returned results.

Daily refund cap (per team, per UTC day, default 100 credits). Once a team's creditsRefundedToday reaches dailyRefundCap, further submissions still record feedback but no longer refund credits. The response sets dailyCapReached: true. Agents should stop calling this tool for the rest of the UTC day when they see that flag.

Usage Example:

{
  "name": "firecrawl_search_feedback",
  "arguments": {
    "searchId": "0193f6c5-1234-7890-abcd-1234567890ab",
    "rating": "good",
    "valuableResults": [
      { "source": "web", "position": 1 },
      { "source": "web", "position": 3 },
      { "source": "news", "position": 2, "reason": "Broke the pricing change." }
    ],
    "valuableSources": [
      {
        "url": "https://docs.firecrawl.dev/pricing",
        "reason": "Not in the results; found via a result's pricing link. Answered the question."
      }
    ],
    "missingContent": [
      {
        "topic": "Pricing for the search endpoint",
        "description": "No pricing tier table for /search specifically."
      },
      { "topic": "Per-team rate limits" }
    ],
    "querySuggestions": "Boost docs.firecrawl.dev for queries that mention 'firecrawl'"
  }
}

Returns:

• { success, feedbackId, creditsRefunded, alreadySubmitted? } JSON.

Sends structured feedback for a completed v2 endpoint job through /v2/feedback. Use this for endpoint-level feedback on scrape, parse, map, or search jobs. For search-result quality specifically, prefer firecrawl_search_feedback because it includes search-specific guidance.

Keep feedback concise: use issue codes, tags, short notes, URLs, page numbers, and small metadata objects. Do not include raw scrape/parse outputs.

Opt out: set FIRECRAWL_NO_ENDPOINT_FEEDBACK=1 (or FIRECRAWL_DISABLE_ENDPOINT_FEEDBACK=1) in the environment when starting the MCP server. The firecrawl_feedback tool will not be registered, so agents cannot call it.

Usage Example:

{
  "name": "firecrawl_feedback",
  "arguments": {
    "endpoint": "scrape",
    "jobId": "0193f6c5-1234-7890-abcd-1234567890ab",
    "rating": "partial",
    "issues": ["missing_markdown"],
    "tags": ["docs"],
    "note": "The pricing table was missing from the markdown output.",
    "url": "https://example.com/pricing",
    "pageNumbers": [1],
    "metadata": {
      "format": "markdown"
    }
  }
}

Returns:

• { success, feedbackId, creditsRefunded, creditsRefundedToday?, dailyRefundCap?, dailyCapReached?, websiteCapReached?, alreadySubmitted?, warning? } JSON.

Starts a crawl job, polls until it reaches a terminal state, and returns the final crawl status/data.

Best for:

• Extracting content from multiple related pages, when you need comprehensive coverage.

Not recommended for:

• Extracting content from a single page (use scrape)
• When token limits are a concern (use map + scrape for tighter control)
• When you need fast results (crawling can be slow)

Warning: Crawl responses can be very large and may exceed token limits. Limit the crawl depth and number of pages, or use map + scrape for tighter control.

Common mistakes:

• Setting limit or maxDiscoveryDepth too high (causes token overflow)
• Using crawl for a single page (use scrape instead)

Prompt Example:

"Get all blog posts from the first two levels of example.com/blog."

Usage Example:

{
  "name": "firecrawl_crawl",
  "arguments": {
    "url": "https://example.com/blog/*",
    "maxDiscoveryDepth": 2,
    "limit": 100,
    "allowExternalLinks": false,
    "deduplicateSimilarURLs": true
  }
}

Returns:

• Final crawl status and data after internal polling, including id, status, completed, total, creditsUsed, expiresAt, next, and data. Use the returned id with firecrawl_check_crawl_status if you need to re-check the job later.

Check the status and results of an existing crawl job by ID.

{
  "name": "firecrawl_check_crawl_status",
  "arguments": {
    "id": "550e8400-e29b-41d4-a716-446655440000"
  }
}

Returns:

• Response includes the status of the crawl job:

Parse local files or hosted upload references with Firecrawl's /v2/parse endpoint.

Best for: PDFs, Word documents, spreadsheets, HTML files, and other documents that need markdown or structured JSON output. The hosted endpoint uses a two-step upload flow; local direct file reads require a self-hosted FIRECRAWL_API_URL.

Not recommended for: Remote URLs (use scrape), multiple files in one call (call parse once per file), or browser-only actions such as screenshots and clicks.

Hosted endpoint: it cannot read your filesystem. Call firecrawl_parse with filePath to receive a short-lived upload command and nextToolCall, upload the file locally, then call firecrawl_parse again with the returned uploadRef. The upload step requires Firecrawl auth or the keyless tier. In local npx firecrawl-mcp mode, direct file parsing currently requires FIRECRAWL_API_URL pointing to a self-hosted Firecrawl API; a plain cloud API-key-only local server cannot read and upload files through this tool.

Usage Example:

{
  "name": "firecrawl_parse",
  "arguments": {
    "filePath": "/absolute/path/to/document.pdf",
    "formats": ["markdown"],
    "parsers": ["pdf"],
    "zeroDataRetention": true
  }
}

Returns: Parsed document content or hosted upload instructions with a nextToolCall.

Call firecrawl_scrape once per URL with formats: ["markdown"], then extract the requested fields from the returned content yourself.

{
  "name": "firecrawl_scrape",
  "arguments": {
    "url": "https://example.com/product",
    "formats": ["markdown"]
  }
}

When the URLs are not known or the data spans several sites, use firecrawl_agent for multi-source research.

Autonomous web research agent that returns structured data when you do not know the URLs or the answer spans several sites. Describe the fields you need, optionally pass a JSON schema and seed URLs, and the agent searches, navigates, reads pages, and returns JSON assembled across sources. Use it for an entity plus its fields, for lists and datasets, and for pages that need navigation to reach the data. For one known URL use firecrawl_scrape with markdown and extract the requested fields yourself.

How it works:

The agent performs web searches, follows links, reads pages, and gathers data autonomously. This runs asynchronously - it returns a job ID immediately, and you poll firecrawl_agent_status to check when complete and retrieve results.

Async workflow:

1. Call firecrawl_agent with your prompt/schema → returns job ID
2. Do other work while the agent researches (can take minutes for complex queries)
3. Poll firecrawl_agent_status with the job ID to check progress
4. When status is "completed", the response includes the extracted data

Best for:

• Complex research tasks where you don't know the exact URLs
• Multi-source data gathering
• Finding information scattered across the web
• Tasks where you can do other work while waiting for results

Not recommended for:

• Simple single-page scraping where you know the URL (use scrape with markdown and extract the requested fields yourself)

Arguments:

• prompt: Natural language description of the data you want (required, max 10,000 characters)
• urls: Optional array of URLs to focus the agent on specific pages
• schema: Optional JSON schema for structured output
• effort: Optional. "low", "medium" or "high" reasoning budget for the agent task.
• maxCredits: Optional positive integer. Spending limit in credits for this run. The API defaults to 2,500 when omitted, and caps a free request at 2,500.
• strictConstrainToURLs: Optional boolean. If true, the agent only visits the URLs in urls.
• threadId: Optional. Continue an existing thread: the threadId from an earlier firecrawl_agent or firecrawl_agent_status result. Omit to start a new thread. On a follow-up, omitted mode, urls, schema and exchange settings carry over from the previous turn.
• mode: Optional. "extract" (default) returns the complete structured result every turn. "chat" lets a follow-up that asks for no new data get a short reply in message instead of a re-run; exchange.requireApproval needs it.
• exchange: Optional. Alexandria provider settings for this turn, forwarded as-is to POST /v2/agent:
  • enabled, toolkits (up to 5 provider slugs), maxCalls (1 to 30), requireApproval (paid calls end the turn with a pendingApproval; needs mode: "chat" on the same call, even on a follow-up)
  • onTermsRequired: what to do when an Alexandria provider the agent would use needs data terms your team has not accepted. Gated providers are never called in any mode. Omitted on a follow-up keeps the previous turn's value.
    • "skip" (default): answer with accepted providers only. exchange.skippedProviders on the status result lists the gated providers that would have helped.
    • "ask": the same, plus a terms pendingApproval and exchange.requiresAction with the approval ID and provider requirements. Read terms with terms/show; an organization admin accepts them in the Firecrawl dashboard.
  • approve: { approvalId, callIds?, always? } answers yes to the pendingApproval the previous turn ended on. callIds and always apply to paid-call approvals only.
  • decline: { approvalId } answers no. A declined terms offer keeps those providers out of the rest of the thread.
  • approve and decline need threadId, and only one of them can be sent.

Provider terms (ask mode): there is no auto-accept mode. When a turn ends on a terms offer, the status result carries pendingApproval (kind: "terms") and exchange.requiresAction with the approvalId and provider requirements. Any terms/accept descriptor in that API payload is unavailable through MCP. To use the provider:

1. Show the user the terms (terms/show through alexandria_library_read).
2. Direct an organization admin to accept the terms at the provider's URL, or data sources settings. A data request is not consent.
3. Wait for the admin to confirm acceptance in the dashboard.
4. Continue the same thread: call firecrawl_agent with the same threadId and exchange.approve: { "approvalId": "..." }. This resumes research and does not accept terms.

If the user says no, call firecrawl_agent with the same threadId and exchange.decline: { "approvalId": "..." } instead.

Prompt Example:

"Find the founders of Firecrawl and their backgrounds"

Usage Example (start agent, then poll for results):

{
  "name": "firecrawl_agent",
  "arguments": {
    "prompt": "Find the top 5 AI startups founded in 2024 and their funding amounts",
    "schema": {
      "type": "object",
      "properties": {
        "startups": {
          "type": "array",
          "items": {
            "type": "object",
            "properties": {
              "name": { "type": "string" },
              "funding": { "type": "string" },
              "founded": { "type": "string" }
            }
          }
        }
      }
    }
  }
}

Then poll with firecrawl_agent_status using the returned job ID.

Usage Example (with URLs - agent focuses on specific pages):

{
  "name": "firecrawl_agent",
  "arguments": {
    "urls": ["https://docs.firecrawl.dev", "https://firecrawl.dev/pricing"],
    "prompt": "Compare the features and pricing information from these pages"
  }
}

Usage Example (follow-up on the same thread):

{
  "name": "firecrawl_agent",
  "arguments": {
    "prompt": "Only keep the startups based in Europe",
    "threadId": "0199a1b2-0000-7000-8000-000000000031"
  }
}

Usage Example (continue the thread after an admin confirmed dashboard acceptance):

{
  "name": "firecrawl_agent",
  "arguments": {
    "prompt": "The admin confirmed acceptance of the Apollo terms in the dashboard. Continue.",
    "threadId": "0199a1b2-0000-7000-8000-000000000031",
    "exchange": { "approve": { "approvalId": "0199a1b2-0000-7000-8000-000000000033" } }
  }
}

Returns:

• Job ID for status checking, plus threadId and threadTurn. Use firecrawl_agent_status to poll for results.

Check the status of an agent job and retrieve results when complete. Use this to poll for results after starting an agent.

Polling pattern: Agent research can take minutes for complex queries. Poll this endpoint periodically (e.g., every 10-30 seconds) until status is "completed" or "failed".

{
  "name": "firecrawl_agent_status",
  "arguments": {
    "id": "550e8400-e29b-41d4-a716-446655440000"
  }
}

Possible statuses:

• processing: Agent is still researching - check back later
• completed: Research finished - response includes the extracted data
• failed: An error occurred

Interact with a fresh URL or with a page that was already opened by firecrawl_scrape.

Best for: Clicking, typing, navigating, and extracting state from dynamic pages without restoring the deprecated browser tools.

Usage options:

• Pass url to scrape and open a page for interaction in one MCP call.
• Pass scrapeId to continue interacting with an existing scraped page.
• To save browser state (cookies, localStorage) to a named profile, pass url with scrapeOptions: { "profile": { "name": "my-profile", "saveChanges": true } }. The state is saved when firecrawl_interact_stop ends the session.
• Pass exactly one of url or scrapeId, plus either prompt or code.

Usage Example:

{
  "name": "firecrawl_interact",
  "arguments": {
    "url": "https://example.com",
    "prompt": "Click the pricing link and summarize the visible plans"
  }
}

Returns: Interaction result and, for URL mode, the derived scrapeId for follow-up or cleanup.

Stop an interact session for a scraped page when you are done interacting.

{
  "name": "firecrawl_interact_stop",
  "arguments": {
    "scrapeId": "scrape-id-here"
  }
}

Search and inspect papers through the research MCP tools. For GitHub repositories, issues, and pull requests, use firecrawl_developer_search.

Covers: paper abstracts and full text across biomedical, life-science, and clinical literature (PubMed, bioRxiv, medRxiv) alongside arXiv and other scientific sources.

Available research tools:

• firecrawl_research_search_papers: search paper metadata and abstracts with a natural-language query, with optional author, category, and date filters.
• firecrawl_research_inspect_paper: retrieve canonical metadata for one paper ID (arXiv, PMC, PMID, or DOI).
• firecrawl_research_related_papers: expand from one or more anchor papers through the citation graph.
• firecrawl_research_read_paper: read full-text passages from a specific paper.

Best for: Literature review, paper lookup, and repository discovery workflows where the agent needs a focused research surface instead of general web search.

firecrawl_search with categories: ["research"] is a different surface: it filters ordinary web results to research-affiliated websites and returns page snippets, not paper records. Use these tools when the question is about the literature itself, and pass several distinct framings of the same question, since they surface different papers than a single query does.

Create and manage recurring page monitors. Monitors run scheduled scrapes or crawls, diff each result against the last retained snapshot, and can notify by webhook or email.

Best for:

• Watching one page or a few pages over time
• Alerting on meaningful changes using a plain-English goal
• Tracking check history and page-level diffs

Recommended create pattern:

Use page or pages plus goal. The MCP server builds the monitor request with a 30-minute schedule and the API enables meaningful-change judging automatically.

Meaningful-change judging runs automatically when goal is set. Page webhooks expose isMeaningful and judgment on monitor.page events.

Write goals as concise 2-3 sentence monitor instructions. Say what should trigger an alert, preserve any scope the user gave, and include intent-specific exclusions only when obvious from the request. Generic noise such as whitespace, formatting-only changes, request IDs, tracking params, generic metadata, and unrelated page chrome is already handled by the judge, so do not repeat it in every goal. If the user is vague, keep the goal broad; if they ask for broad monitoring or "any change", preserve that. If the user says they do not care about something, include that explicitly.

{
  "name": "firecrawl_monitor_create",
  "arguments": {
    "page": "https://example.com/pricing",
    "goal": "Alert when pricing, packaging, or launch messaging changes."
  }
}

Multiple pages with webhooks:

{
  "name": "firecrawl_monitor_create",
  "arguments": {
    "pages": ["https://example.com/pricing", "https://example.com/changelog"],
    "goal": "Alert when pricing, packaging, or launch messaging changes.",
    "webhookUrl": "https://example.com/webhooks/firecrawl"
  }
}

Advanced create requests:

Pass body when you need crawl targets, JSON change tracking, custom retention, or explicit judgeEnabled control.

{
  "name": "firecrawl_monitor_create",
  "arguments": {
    "body": {
      "name": "Docs monitor",
      "schedule": { "text": "hourly", "timezone": "UTC" },
      "goal": "Alert when docs pages add, remove, or materially change API behavior.",
      "targets": [{ "type": "crawl", "url": "https://example.com/docs" }]
    }
  }
}

Other monitor tools:

• firecrawl_monitor_list: list monitors.
• firecrawl_monitor_get: get one monitor.
• firecrawl_monitor_update: update fields including goal, judgeEnabled, webhook, and notification.
• firecrawl_monitor_run: trigger a check now.
• firecrawl_monitor_delete: delete a monitor (destructive; only call when the user intends to remove it).
• firecrawl_monitor_checks: list checks, optionally filtered by status.
• firecrawl_monitor_check: get page-level results, including diff, snapshot, judgment.meaningful, and judgment.meaningfulChanges.

Search an index built for coding agents. The index covers GitHub issues, merged pull requests, repository READMEs, and code documentation.

Best for: A programming question: code behaviour, a library or framework, an API contract, an error message, or a known bug.

Arguments:

{
  "name": "firecrawl_developer_search",
  "arguments": {
    "query": "how do I configure retries",
    "k": 10,
    "skills": "only"
  }
}

• query (required): the developer question or search phrase.
• k: number of ranked results. The default is 10 and the maximum is 100.
• skills: set to "only" to search agent-skill files alone.

Returns: Ranked results. Each result carries an ID, a source type (issue, pull_request, readme, or doc), a URL, a title, and the matched passages in markdown.

firecrawl_search with categories: ["developer"] searches the same index beside the web results. Use this tool instead when you want the matched passages, the skills filter, or no web results in the response. The search-only endpoint exposes both tools, and the same choice applies there.

Search primary law and regulatory material from US federal, state, and local government sources: statutes, regulations, codes, court opinions, and other government publications.

Best for: A legal or regulatory question that needs the governing text or an official government source.

Arguments:

{
  "name": "firecrawl_gov_search",
  "arguments": {
    "query": "excessive force claims against police officers",
    "k": 10,
    "courtType": "federal-appellate",
    "cites": "42 U.S.C. § 1983"
  }
}

• query (required): the legal or regulatory question or search phrase.
• k: number of ranked results. The default is 10 and the maximum is 100.
• jurisdiction: keep only pages of one jurisdiction: "us" for federal, or "us-" plus a two-letter state or territory code, such as "us-ca" or "us-dc".
• court: keep only pages labeled for one court, by CourtListener court id, such as "scotus", "ca9" or "nysd".
• courtType: keep only pages from one type of court: federal-supreme, federal-appellate, federal-district, federal-bankruptcy, federal-special, military, state-supreme, state-appellate, state-trial, state-special or state-attorney-general.
• cites: keep only pages that cite one law or case, as a printed citation such as "42 U.S.C. § 1983" or "5 U.S. 137", or a citation id such as "us:usc:42:1983".
• domain: keep only pages from one site, such as "irs.gov".

The filters are optional and combine: a result must match every filter set. Filters keep only pages the index has labeled, so a filtered search can leave out relevant pages, and no results does not mean none exist. Court labels are most complete for federal courts and case-law collections.

Returns: Ranked results. Each result carries a position, a title, a URL, and the matched snippet. The search-only endpoint exposes this tool too.

firecrawl_search with categories: ["gov"] searches the same sources and returns the hits in data.web with category: "gov". The gov category does not take these filters and cannot be combined with other categories.

Firecrawl Alexandria is a catalogue of data providers reachable through the Firecrawl API with a Firecrawl API key on a team with Alexandria access. Keyless sessions (hosted or local) get Alexandria requires an API key on a team with Alexandria access; Alexandria discovery tools are not listed for hosted keyless sessions. A stdio server pointed at its own API with FIRECRAWL_API_URL is the exception: it sends Alexandria tools to that API, which answers for the account.

Semantic discovery (firecrawl_search):

{
  "name": "firecrawl_search",
  "arguments": {
    "query": "podcast conversations about AI agents",
    "sources": ["web", "alexandria"],
    "domainTools": true,
    "limit": 2
  }
}

data.tools defaults to compact suggestions with provider, capability, and description. Set toolDetail: "summary" for metadata and navigation, or toolDetail: "full" for contracts including inputs, response fields and examples. domainTools: true adds contextual matches to query mentions and result URLs in that same array. Check warning for unavailable discovery. Search requires a query and does not accept catalogue traversal filters. This discovery works on both the full and search-only MCP surfaces.

Explore the library (alexandria_library_search): the catalogue is a read-only filesystem, and the tool runs one shell command over it.

{
  "name": "alexandria_library_search",
  "arguments": { "command": "grep -ril \"podcast\" /library | head -20" }
}

A file path is the execute address: /library/finance/fred/series/observations.md is provider fred, capability series/observations. Typical commands:

Commands are ls, tree, cat, head, tail, grep, find, wc, sort, uniq, cut, tr, sed, jq and xargs, with pipes. There is no network and the library cannot be written. The result is the command's output as text; structuredContent carries stdout, stderr, exitCode, and truncated when output was cut at 30,000 characters. A command that finds nothing is not an error: grep exits 1 and the text is [no output; exit code 1]. Library lookups cost zero credits and never run a provider. The files hold what the same account would see in discovery, without prices.

This tool requires the Alexandria API deployment that serves firecrawl/library.

Execute (alexandria_library_read): calls is an array of one to ten {provider, capability, options} entries, which may name different providers. provider and capability come from search results or a library path, and options follow the contract. requestId and timeout may also be supplied. Results come back in data.alexandria in the same order, and one call can fail while the others succeed.

{
  "name": "alexandria_library_read",
  "arguments": {
    "calls": [
      { "provider": "fred", "capability": "series/observations", "options": { "series_id": "CPIAUCSL" } },
      { "provider": "fred", "capability": "series/observations", "options": { "series_id": "MORTGAGE30US" } }
    ]
  }
}

One call sent as the arguments themselves ({provider, capability, options}) is read as a calls array of one.

Earlier tools (firecrawl_find_tools, and firecrawl_scrape with alexandria): this server still lists both beside the two new tools, with the definitions they had before them, for clients that take up a new or changed tool only after their own review. Both answer the calls they used to, so clients and prompts that learned them keep working; new ones should use alexandria_library_search and alexandria_library_read. firecrawl_find_tools browses the catalogue ({} for categories, then categories, providers and capabilities to narrow to one contract, or query and urls for lookup), and firecrawl_scrape takes alexandria in place of url, as one {provider, capability, options} call or an array of up to ten, with requestId and timeout.

The search-only endpoint lists and documents the earlier tools, because its tool list is fixed by its connector listing. An embedder whose toolFilter leaves either new tool out gets the earlier tools too.

Returns: { success, scrape_id, requestId, data: { alexandria: [...], creditsCost } }. Each item is either a result (provider, capability, data, records, upstreamStatus) or an error with a code; a provider error is reported in its item, not as a failed call, and data.creditsCost is the total charged for the request. Alexandria error bodies (403 without Alexandria access, and 402/409 billing statuses) are relayed in-band with their code.

Execution generates one x-request-id and returns it on success or failure. Retry the identical payload with that requestId; do not create a new ID after an uncertain outcome. Available credits are checked by the API.

An Alexandria provider whose terms the team has not accepted fails before anything runs with HTTP 403 and this body:

{
  "success": false,
  "code": "THIRD_PARTY_DATA_TERMS_REQUIRED",
  "error": "An organization admin must accept the benzinga provider's terms (version 2026-09-12-placeholder) before this request can run. Accept them at https://www.firecrawl.dev/app/alexandria/benzinga",
  "requiresAction": {
    "type": "accept_terms",
    "terms": "benzinga",
    "version": "2026-09-12-placeholder",
    "url": "https://www.firecrawl.dev/app/alexandria/benzinga"
  }
}

The tool result relays it as an error with structuredContent carrying code, status: 403, requestId, the requiresAction object unchanged, and next_actions (human_action_required then retry_same_request). Accepting terms is a legal act, so an organization admin accepts them in the Firecrawl dashboard, not through MCP. Use the returned nextTool call to read the agreement: alexandria_library_read with calls: [{provider: "firecrawl", capability: "terms/show", options: {provider: "<provider>"}}] (on the search-only endpoint, the same call through firecrawl_scrape with alexandria), sent separately from provider execution, and present it to the user. An organization admin then accepts it at requiresAction.url, or at https://www.firecrawl.dev/app/settings?tab=data-sources. Both tools refuse every other terms/* capability, so they make no account changes. A request for data is not consent, and no automatic acceptance or uncertain retries occur. No credits are charged for the blocked retrieval. After the admin confirms acceptance, call the same tool again with the identical payload and requestId.

People and company search and enrichment over Alexandria's B2B data providers. Both tools are listed wherever the server has a credential: they are not listed in local keyless mode or to keyless sessions. The search-only endpoint includes them too. They need a Firecrawl credential (an API key or OAuth access) on a team with Alexandria access.

people_company_search takes one normalized input and sends it to Alexandria's search waterfall, firecrawl/search/auto, as a single request: entity ("person" or "company", required), optional query, filters (titles, seniority, functions, company_domains, locations, industries, technologies, keywords, and headcount and founded_year as { "min", "max" }), limit (1 to 100, default 10), cursor, count_only and max_credits (default 1000). For people, company filters describe the employer. The waterfall asks its providers one at a time in list-price order, and only those that apply every filter given; when none does, it searches companies first and then people at the domains it found, and a request no provider or chain covers is refused with unsupported_filters, naming the filter to drop. With company_domains it stops at the first provider that returns anyone; otherwise it fills limit across up to three providers, asking each only for the rows still missing, and drops duplicates. A row whose own title, employer domain, country or city, headcount or founding year fails a filter given, or whose title names another seniority level or function than those given (SVP, EVP and AVP meet vp), is dropped, and the search keeps filling from the next provider. Rows carry name, title (title_match is current or past), company, location, LinkedIn URL and source, and never emails or phones. A request that names one person is refused with use_enrich and pointed to people_company_enrich. providers lists each provider call with its status, rows, rows dropped for failing a filter, rows kept as unverified because their title names no level or function for a seniority or functions filter, credits and time. provider limits the call to one provider, and with options sends that provider's own search options as given.

{
  "name": "people_company_search",
  "arguments": { "entity": "person", "filters": { "titles": ["Head of Data"], "company_domains": ["example.com"] }, "limit": 10 }
}

people_company_enrich takes records, not provider blocks, and runs Firecrawl's enrichment waterfall on them. people takes name, first_name, last_name, company, domain, linkedin_url, email and phone; each person needs a linkedin_url, email or phone, or a name with a company or domain. companies takes domain, name and a company linkedin_url, and each needs one of them. A call takes at most 10 records, people and companies together.

Without contact_fields each person gets a profile and each company gets firmographics. contact_fields (work_email, personal_email, mobile) asks for those values for each person, together with the person's profile, found first so the contact lookups can use it (profile: false skips it); companies in the same call still get firmographics, and contact_fields with only companies fails with invalid_input. Mobile is opt-in: phone providers, the costliest lookups, run and bill only when contact_fields lists mobile.

For each requested field, Firecrawl tries the providers that can use the record's inputs one at a time, ordered by listed price and past success rate, and stops at the first one that finds the field. A provider that has no match, errors or is rate limited is skipped for the next one, without retries. A person known only by an email or phone is first resolved to a profile, and the contact providers then use what it revealed.

{
  "name": "people_company_enrich",
  "arguments": {
    "contact_fields": ["work_email"],
    "people": [{ "name": "Jane Doe", "domain": "example.com" }, { "linkedin_url": "https://www.linkedin.com/in/janedoe" }],
    "companies": [{ "domain": "example.com" }]
  }
}

The response has people and companies lists with one entry per input, each with input_index and results: one { "provider", "data" } entry per provider that found a requested field, where data is that provider's own response, unchanged. Each record also has fields: for each filled field, the provider behind it; for an email the chosen value, verified (the provider marked it deliverable) and domain_match; for a profile or firmographics complete, false when only a partial answer was found, and for firmographics linkedin_confirmed. A work email is never a free-mail or role address, and is on the company's domain when one is known; verified_only: true returns only verified emails. A record nothing was found for has empty results. credits_used is what the call charged: each provider that found something is charged its listed price, and misses are free. max_credits (at most 10,000) caps the call: a provider is tried only while its listed price still fits. Without it the call is uncapped. stopped: "team_quota" means the team ran out of credits during the call. To call one specific provider, use firecrawl_scrape with an alexandria body.

Both tools also accept, and drop, the conversationId and objective fields agents carry between Firecrawl tools.

The tool requires an authenticated Firecrawl account and is read-only.

• firecrawl_credit_usage defaults to { "view": "current" } and returns remainingCredits, planCredits, billingPeriodStart, and billingPeriodEnd. Remaining credits can exceed plan credits when the team has top-ups or grants.
• Pass { "view": "historical" } for calendar-month periods containing startDate, endDate, and creditsUsed. Passing { "byApiKey": true } also selects the historical view and splits periods by API key; each row then includes apiKey. Passing { "byEndpoint": true } instead splits periods by the endpoint that used the credits (scrape, search, crawl, map, agent, ...); each row then includes endpoint. Pass { "bin": "day" } for one period per UTC day instead of per calendar month. Do not combine byApiKey with byEndpoint, or any of these with { "view": "current" }. The latest period's endDate can be null. If the Firecrawl API does not offer the endpoint split or daily periods yet, the result carries a notice and monthly figures.

The package also exposes a library entry point for embedding the server in another runtime. It is unstable: the API, its options and its types may change in any release, including minor and patch releases, so pin an exact version.

npm install --save-exact firecrawl-mcp zod

The example defines tool parameters with zod, so the embedding project installs it as its own dependency.

import { createFirecrawlMcpServer } from 'firecrawl-mcp/server';
import { z } from 'zod';

const server = createFirecrawlMcpServer({
  transport: 'httpStream',
  apiKey: process.env.FIRECRAWL_API_KEY,
  unstable_hooks: {
    registerExtraTools: (registrar) => {
      registrar.addTool({
        name: 'my_tool',
        description: 'An extra tool served next to the Firecrawl tools.',
        parameters: z.object({ text: z.string() }),
        execute: async ({ text }) => text,
      });
    },
  },
});

await server.start({
  transportType: 'httpStream',
  httpStream: { port: 3000, host: '127.0.0.1', endpoint: '/mcp', stateless: true },
});
// MCP endpoint: http://127.0.0.1:3000/mcp

Set host explicitly. Without it the server listens on localhost, which Node binds to a single address (often only ::1), so requests to 127.0.0.1 are refused. Use 0.0.0.0 or :: to accept connections from other machines only behind network access controls or an unstable_hooks.authenticate check: without one, every caller that can reach the port is served with the configured apiKey.

The library doesn't check Host or Origin unless asked, and its CORS headers allow any origin. A server reachable from a browser on the same machine should pass httpStream.allowedHosts (for example ['localhost', '127.0.0.1', '[::1]']): requests whose Host or Origin header names another host then get 403 (except GET and HEAD on the health path), and CORS headers echo the allowed origin. The command line sets this list in local HTTP mode.

Options mirror the CLI's environment variables: apiUrl, apiKey, transport, logging, safeMode, fileAccess (local or upload), requireCredential, searchFeedback and endpointFeedback. unstable_hooks accepts:

The server is built on the official MCP TypeScript SDK (@modelcontextprotocol/server). Use the types exported from firecrawl-mcp/server (Session, ToolDefinition, ToolRegistrar, FirecrawlMcpServerHooks and others) rather than the SDK's own.

The entry point also exports the helpers the built-in tools use, such as recoveryPayload, normalizeHeader, requestOrigin, isRuntimeUserAgent, initializeClientSlug, searchToolBaseFields, searchOutputSchema and structuredCompact, so extra tools and hooks can match the built-in behaviour. They are unstable as well.

The server includes comprehensive logging:

• Operation status and progress
• Performance metrics
• Rate limit tracking
• Error conditions

Example log messages:

[INFO] Firecrawl MCP Server initialized successfully
[INFO] Starting scrape for URL: https://example.com
[ERROR] Rate limit exceeded

The server provides robust error handling:

• API rate-limit errors surfaced to the MCP client
• Detailed error messages
• Network resilience

Example error response:

{
  "content": [
    {
      "type": "text",
      "text": "Error: Rate limit exceeded"
    }
  ],
  "isError": true
}

# Install dependencies
npm install

# Build
npm run build

# Run tests
npm test

1. Fork the repository
2. Create your feature branch
3. Run tests: npm test
4. Submit a pull request

Thanks to @vrknetha, @cawstudios for the initial implementation!

Thanks to MCP.so and Klavis AI for hosting and @gstarwd, @xiangkaiz and @zihaolin96 for integrating our server.

MIT License - see LICENSE file for details

Authenticated search defaults to web results, semantic Alexandria tools, and domain matches. Start with the actual question. Use alexandria_library_search to check whether a provider covers the data and to read a selected contract: the library README, a category, a provider, then one capability file. Run capabilities through alexandria_library_read; URL scraping and search never execute provider tools. The search-only endpoint has firecrawl_find_tools and firecrawl_scrape with alexandria for these two steps.

Read the recorded example under /examples only if the input or output shape is unclear. Inspect related capabilities together, for example cat two contracts in one command, and reuse what you read.

For a potentially large workflow result, supply and preserve a top-level requestId before execution. That ID remains available even if the client rejects the response. Regular URL scrapes use the returned scrape ID instead.

For a large retained workflow or regular scrape result, call alexandria_library_read with:

{
  "calls": [
    {
      "provider": "firecrawl",
      "capability": "bash",
      "options": {
        "requestId": "<source-request-or-scrape-id>",
        "command": "ls -lh"
      }
    }
  ]
}

Read stdout, stderr, exitCode, and workspaceId in data.alexandria[0].data. Continue with options: {workspaceId, command} to inspect response.json using jq, or document.md using bounded text commands for regular scrapes. Keep output selective. Source loading must be a standalone call; its nested source ID differs from the top-level execution request ID. Workspaces expire after five idle minutes, and not every result is retained (including ZDR and API-provider workflow payloads). Search IDs are not supported.

For workflow sources, response.json preserves the API envelope: select .data.alexandria[].data, then the selected contract’s response.key when nonempty. Combine related counts and projections in one Bash command when that shape is known, rather than repeatedly inspecting keys.

Successful Alexandria execution responses above 20,000 estimated tokens (serialized UTF-8 bytes divided by four) return a small handoff after remote Bash confirms the complete batch is accessible. Follow nextTool to inspect the retained data. This adds one Bash call and starts a workspace with a five-minute idle TTL. The full payload is preserved. If confirmation fails, the original response stays inline. Search, ordinary URL scrape, error responses and Firecrawl utility calls are unchanged. This delivery budget does not measure the client’s remaining context; no process-local result cache is added. When local filesystem tools are available, saving CLI output and reading selected sections is another option.

The CLI discovery sequence maps to these MCP calls:

The search surface keeps firecrawl_find_tools and firecrawl_scrape with alexandria for the last five rows. Reuse a complete contract from search when present rather than making another lookup.

The existing firecrawl_feedback tool accepts endpoint: "alexandria":

{
  "endpoint": "alexandria",
  "rating": "partial",
  "requestedWebsite": {
    "url": "https://example.com",
    "requestedFunctionality": "Find records and download their attachments"
  },
  "objective": "Compare contract requirements across agencies before bidding",
  "rationale": "Found summaries but could not retrieve attachments"
}

The optional objective is the underlying goal of the session: what the agent or its user was ultimately trying to accomplish, beyond the single website in requestedWebsite.

This uses authenticated POST /v2/feedback, without a job ID. It must arrive within 20 minutes of the team's most recent Alexandria search, discovery, or execution; later feedback is rejected with FEEDBACK_WINDOW_EXPIRED. Eligible feedback can refund 1 credit, subject to daily caps of 10 credits per website and 100 credits per team (UTC day). Past either cap, feedback is still recorded with no refund, and the response sets websiteCapReached or dailyCapReached. Optional providerFeedback and capabilityFeedback arrays describe coverage gaps and execution issues; the tool schema lists supported issue values. A new_capability_request requires requestedFunctionality; missing_capability (the provider exists but lacks the capability) does not. Existing feedback opt-out and authentication controls apply.

Alexandria provider call results can carry a feedbackToken so ratings are attributed to the calls they are about. The server keeps each token per MCP session (keyed by the Mcp-Session-Id and the caller's credential, or by the credential alone when the client sends no session id; at most 20 tokens, dropped after 24 hours) and shows the agent a short feedbackRef such as c3 in its place, so long tokens never reach the model. firecrawl_feedback sends the session's tokens with every Alexandria rating, so the overall rating covers every retained tokenized call in the session, and accepts an optional callRatings array to rate individual calls by their ref:

{
  "endpoint": "alexandria",
  "rating": "partial",
  "requestedWebsite": { "url": "https://example.com", "requestedFunctionality": "Find records" },
  "rationale": "The search call worked, the detail call failed",
  "callRatings": [
    { "ref": "c1", "rating": "good" },
    { "ref": "c2", "rating": "bad" }
  ]
}

Refs that are unknown or expired are ignored, and the result says so in note. A successful submission clears the tokens it sent, so the next rating covers only new calls. Results that carry no token (for example under zero data retention) keep working as before.

Eligible Alexandria execution and discovery results include a feedbackTool pointer with the tool name and a skeleton of the arguments. The pointer is omitted for Firecrawl-internal calls such as bash and when firecrawl_feedback is not registered (FIRECRAWL_NO_ENDPOINT_FEEDBACK or keyless startup).