Programmatic access and automation
Overview
There are three ways to work with Whisperer without using the interface: the MCP server for AI clients, connected services for routine automation, and export for moving material elsewhere. Whisperer does not publish a general-purpose REST API — the addresses under /v1/ are the private channel between our own apps and the backend, they carry no compatibility promise, and they may change at any time.
That is a deliberate boundary, not an omission. A published API has to keep working for years; MCP gives the same access with versioning, consent screens and revocation already built in.
When to use it
- You want an AI assistant to read your meetings and act on your tasks — use MCP.
- You want a repeating action ("new meeting → row in a spreadsheet") — use Zapier.
- You want the material itself in another tool — use export.
Step-by-step
- For AI clients, connect over MCP: see MCP: connecting Claude, ChatGPT and Cursor. This is the supported programmatic route and the only one covered by a stability promise.
- For automation, connect Zapier, Notion, Slack or HubSpot in your account, then build the scenario in that service.
- For one-off transfers, export a meeting to Markdown, SRT, VTT or JSON from the meeting page.
- For machine-readable product information — for example when you want an AI assistant to answer questions about Whisperer itself — point it at
https://whisperer.one/llms.txtor the full corpus athttps://whisperer.one/llms-full.txt. Both are public and need no authentication.
What is public
| Address | What it is |
|---|---|
https://whisperer.one/mcp |
MCP server for external AI clients |
https://whisperer.one/.well-known/oauth-protected-resource |
Where a client learns how to authenticate |
https://whisperer.one/.well-known/oauth-authorization-server |
OAuth endpoints and supported scopes |
https://whisperer.one/llms.txt |
Index of the product for AI systems |
https://whisperer.one/llms-full.txt |
The whole public documentation in one file |
https://whisperer.one/docs/en/<page>.md |
Any English documentation page as clean Markdown |
Common mistakes
- Building on
/v1/. Those addresses are internal. They are closed to crawlers, their schema is switched off in production, and they change without notice. - Looking for an API key. There is none, because there is no public REST API. External access is a bearer token issued through the MCP consent flow.
- Expecting webhooks. Outgoing automation goes through Zapier and connected services rather than webhooks of our own.
Best practices
- Read the OAuth endpoints from the metadata document rather than hard-coding them: they are published there precisely so they can move without breaking clients.
- Give each integration its own connection so you can revoke one without affecting the rest.