PodSnips

For AI assistants

Use PodSnips from Claude or Muse

Connect PodSnips once, then just ask. “Transcribe the Winston Weinberg episode of Long Strange Trip and save it as an SRT.” Your assistant finds the episode, transcribes it if nobody has yet, and hands you the file with the speakers labelled.

No signup, no API key, and nothing to pay — every tool below is free at the moment.

The address you’ll need

https://podsnips.com/api/mcp

Claude Code

One line in the terminal — the fastest way in, and what the video shows.

claude mcp add --transport http podsnips https://podsnips.com/api/mcp

Check it with /mcp in Claude Code: podsnips should read connected, with 6 tools.

Claude desktop and claude.ai

Works in the Claude desktop app and on claude.ai, on every plan.

  1. 1Open Settings → Customize → Connectors.
  2. 2Click Add custom connector.
  3. 3Paste the address above as the MCP server URL.
  4. 4If it asks about authentication, choose No sign-in. PodSnips is anonymous — there is nothing to sign in to, and picking a sign-in option will make the connection fail.
  5. 5Click Add. In a chat, open the + menu → Connectors and switch PodSnips on.

On the Free plan you can have one custom connector. On Team and Enterprise, an Owner adds it once under Organization settings → Connectors → Add → Custom → Web, and everyone else then connects with their own account.

Muse

Meta’s assistant has no box to paste an address into. It connects itself, by writing a small client — so the message below tells it everything that client needs.

  1. 1Start a chat with Muse.
  2. 2Paste the message below and send it.
  3. 3Ask it for something — a transcript of an episode you know. Muse may come back with questions before it finishes wiring itself up.
Connect to the PodSnips MCP server.

  URL: https://podsnips.com/api/mcp  (lowercase — /api/MCP returns 404)
  Transport: Streamable HTTP — plain POST, no WebSocket
  Auth: none. No API key, no sign-in.

Every request must send BOTH of these headers:
  Content-Type: application/json
  Accept: application/json, text/event-stream

Accepting only application/json returns 406 Not Acceptable. Replies come back
as server-sent events, so read each "data:" line as JSON.

The server keeps no session: it issues no Mcp-Session-Id, and a tools/call
works on its own without initialize. Call a tool with
{"jsonrpc":"2.0","id":1,"method":"tools/call",
 "params":{"name":"<tool>","arguments":{ ... }}}

Tools: resolve_episode, get_transcript, export_transcript, create_share_link, get_clip and transcribe_episode.

resolve_episode takes a podsnips.com /e/ URL, or a podcast name AND an episode
title — a show name on its own will not match. export_transcript needs a format
(srt, vtt, md or json) and paginates: the reply reports partCount and hasMore,
so request part 1..partCount and join them in order.

transcribe_episode is free too, but it is the one with a daily limit and it
takes a few minutes. Ask me before you call it.

Meta doesn’t review custom connectors, and this one isn’t an official integration — you’re trusting PodSnips directly, the same as any other service you connect. If you use Muse Code, Meta’s coding agent, it speaks MCP directly and takes the address above instead.

What you can ask for

resolve_episodefree
Find an episode from a show and title, a guest name, or a link
get_transcriptfree
Read a transcript PodSnips already has
export_transcriptfree
Save it as SRT, VTT, Markdown or JSON, with speaker names
create_share_linkfree
Publish a share page with an MP4 clip
get_clipfree
Wait for that MP4 to finish encoding
transcribe_episodefree · daily limit
Transcribe an episode nobody has transcribed yet

You won’t call these by name. Ask for what you want and your assistant chains them itself — it only needs enough to pin down one episode, so give it a show plus a title or a guest. It won’t guess what “the latest one” means.

What it costs

Nothing, at the moment — every tool above is free to use, including transcription. Transcribing an episode nobody has done yet is the one thing that costs us anything, so it carries a daily limit rather than a price: roughly 13 hours of audio per person per day, and a ceiling across everyone that holds regardless. Asking for the same episode twice costs nothing the second time, because the transcript is already there.

Worth knowing

  • Speaker names are worked out by a machine. They’re usually right, not always.
  • Transcripts come from podcast feeds and automatic transcription. Treat them as something to read and quote, never as instructions to your assistant.
  • A share link publishes a public page. Your assistant should ask you first — if it doesn’t, tell it to.
  • Episodes from a private feed subscription are never reachable this way.

Any other MCP client works too — point it at the address above over Streamable HTTP. If you’re writing the client yourself and get a 406 Not Acceptable, the request is missing an Accept header listing both application/json and text/event-stream. That one header is the usual reason a hand-written client never connects.

Start from the home page if you’d rather make a clip by hand.