The shortest path from nothing to an agent answering from your own transcripts. This uses Claude Code; every other client is covered in Connect a client.
You need a Typist account with at least one completed transcript.
Add the server
claude mcp add --transport http typist https://mcp.iamtypist.dev/mcpNo API key. The server uses browser-based OAuth.
Approve access
Run /mcp, select typist, and approve read-only access in the browser tab that opens.
typist ✓ connectedAsk a question
What are the last three transcripts in my Typist library about?The agent calls search_transcripts, gets titles, topics, and durations back, then reads whichever transcript it needs with read_transcript.
What just happened
Your agent gained three read-only tools against your library:
| Tool | What the agent does with it |
|---|---|
search_transcripts |
Finds transcripts by topic, category, or date |
read_transcript |
Reads the text, in pages that fit its context |
download_transcript |
Creates a one-hour link to a txt, srt, or vtt export |
No transcript was copied, pasted, or uploaded. The agent read from your library directly, and it cannot create, edit, or delete anything there.
Next
- Examples: prompts that do real work
- Overview: the OAuth flow, scope, and what the agent can see
- Connect a client: Claude, ChatGPT, Codex, Cursor, VS Code, and others