Debriefer
Clients

ChatGPT

Connect ChatGPT to a hosted Debriefer MCP server over HTTP.

ChatGPT talks to Debriefer over HTTP via the Connectors surface. You run the MCP server somewhere it can reach (your own host, Fly, Render, etc.) and register the URL in ChatGPT's settings.

Install

Add to Settings → Connectors → Add connector
ChatGPT connects to the MCP server over HTTP. Run the server with `debriefer-mcp serve --port=8080` and expose it at a public URL.
{  "url": "https://your-mcp-server.example.com/mcp",  "authorization": "Bearer sk_live_..."}
After adding the connector, start a new ChatGPT conversation to see Debriefer tools appear.

Verify it's connected

Start a new conversation in ChatGPT and open the tool tray. The Debriefer tools should appear alongside any other connectors you've enabled. If they don't, see Troubleshooting.

Try a prompt

Say this to ChatGPT

List all my published blueprints.

Client-specific notes

  • One key per connector: ChatGPT sends a single API key with every request to the connector — there's no per-user credential exchange. To isolate access by user, register the connector inside an Enterprise workspace so each member's invocations are attributable through the workspace audit log.
  • Public HTTPS required: The connector URL must be reachable from ChatGPT's servers over HTTPS. Local tunnels (ngrok, Cloudflare Tunnel) work for testing; for anything ongoing, use a stable public host.
  • Health checks: Point your platform's availability probe at /health on the MCP server — ChatGPT will surface a clear error in the connector list if the endpoint stops responding, which makes outages much easier to spot.
  • Hosting options: Fly, Render, Railway, and a plain VM behind a reverse proxy all work. The server is a standard HTTP process, so any platform that can run a Node.js service will do.
  • Rotating the key: Update the key in the connector settings — there's no need to redeploy the server, since the key is sent on each request.

On this page