Connecting

Debriefer runs the MCP server for you. You don't install anything or manage a key - you point your AI assistant at a URL and sign in.

The endpoint

https://mcp.debriefer.ai/mcp

That's the production server. There's also a staging endpoint at https://mcp.debriefer.dev/mcp if you're testing against the preview environment.

It speaks MCP over Streamable HTTP. Any client that can add a remote MCP server by URL and complete an OAuth sign-in can use it.

Authentication: OAuth, not API keys

The server authenticates with OAuth 2.0, the same flow you've used to "Sign in with..." anywhere on the web:

1. You add the URL to your AI assistant.
2. A browser window opens asking you to authorise Debriefer.
3. You sign in to your Debriefer account and approve access.
4. The assistant gets a token and connects. Every Debriefer tool runs as you, in your organisation.

There are no API keys to copy, paste, or rotate. Your assistant holds a short-lived token and refreshes it automatically. To revoke access, remove the connector in your assistant or revoke the session from your Debriefer account.

Under the hood this is standard MCP OAuth - dynamic client registration, PKCE, and a mcp:full scope - so compliant clients discover and complete the flow on their own. You don't need to configure any of it.

What you can connect with

Right now, Claude (web and desktop) is the client we've tested end to end. Adding the URL as a custom connector and authorising in the browser just works.

Other assistants that support remote MCP servers with OAuth should work too, but we haven't verified each one yet. If you get Debriefer running in another client, let us know and we'll add it here.

Next

• Connect Claude - step by step.
• Create your first interview - in under two minutes.
• Building an agent that calls the server programmatically? See Let an agent run an interview.