What this is

Python Bytes runs a public, read-only MCP server. Point Claude, Claude Code, Cursor, or any other MCP-capable assistant at it and it can search the archive, pull up a specific episode's show notes and links, read full transcripts, and check when the next live stream is.

There is no account, no API key, and no rate limit. Everything served here is the same public podcast data already on the website. The server is read-only: nothing an assistant does through it can change anything.

Add it to your assistant

Most clients read a JSON config file. Add the Python Bytes entry to yours and restart the client:

{
  "mcpServers": {
    "python-bytes": {
      "url": "https://pythonbytes.fm/api/mcp"
    }
  }
}

On macOS, Claude Desktop keeps that file at ~/Library/Application Support/Claude/claude_desktop_config.json. For Claude Code, run claude mcp add --transport http python-bytes https://pythonbytes.fm/api/mcp instead. Client setup moves around, so check your client's own docs if the path above does not match what you see.

No MCP support in your tool? Every episode is also available as plain Markdown by adding .md to its URL, and /llms.txt describes the whole site for assistants that only fetch URLs.

Server details

Server name	python-bytes-mcp
Version	1.2.0
Endpoint	https://pythonbytes.fm/api/mcp
Discovery	https://pythonbytes.fm/.well-known/mcp.json
Protocol	JSON-RPC 2.0 over HTTP POST
Authentication	None. Public, read-only data.

Available tools

The server exposes 12 tools. Your assistant discovers them automatically, so this list is here for humans deciding whether it is worth adding.

Calling it directly

The endpoint is plain JSON-RPC 2.0 over HTTP POST, so you do not need an MCP client to use it. Start with the handshake:

curl -X POST https://pythonbytes.fm/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize",
       "params":{"clientInfo":{"name":"my-client","version":"1.0"}}}'

List what is available:

curl -X POST https://pythonbytes.fm/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}'

Then call a tool:

curl -X POST https://pythonbytes.fm/api/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call",
       "params":{"name":"search_episodes","arguments":{"query":"uv","limit":5}}}'

Text or JSON

Tools return human-readable Markdown by default, which is what you want when a language model is going to read the result. Add ?format=json to the endpoint URL and every tool returns structured data instead:

curl -X POST "https://pythonbytes.fm/api/mcp?format=json" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":4,"method":"tools/call",
       "params":{"name":"get_episode_topics","arguments":{"show_id":400}}}'

Either way the payload arrives in the MCP content block, so with ?format=json the text field holds a JSON-encoded string you parse yourself. List tools return an array and single-item lookups return an object.

A note on transcripts

get_transcript_for_episode and get_transcript_vtt return the entire transcript in one block. A single Python Bytes transcript runs to roughly 15,000 tokens, so pull one deliberately rather than in a loop. If you are looking for a specific moment, search_transcripts will narrow it down first.

Questions

Something broken, or a tool you wish existed? Email [email protected]. Talk Python runs its own MCP server too, with the podcast archive and the full course catalog.