Register and verify trusted MCP servers

MCP servers that call your tools need a record you can review, verify, and cut off. Authdog now keeps that record in an environment trust store: register a server, verify it, and rotate the keys used to check it.

This is the longer note behind the June 27 changelog entry.

How it works

Each trust entry supports create, update, verify, revoke, and delete. Keys on an entry can be added, rotated, and revoked without deleting the whole record.

Management is available in the console and over the API, so the same trust decision can be made by an operator or by automation. Verification status is a review signal you set after inspecting the server; revoke is the kill switch when that server should stop being trusted.

This store is the control-plane registry later agent-identity work builds on. Registering a server is not the same as publishing an MCP runtime or granting a human session.

Set it up

In an environment that uses MCP:

1. Create a trust entry for the server you intend to allow.
2. Add a verification key and record the material your runtime will present.
3. Mark the entry verified after review.
4. Rotate the key when the server is rebuilt; revoke the key or the entry to cut trust.

Keep unverified entries out of production traffic. Status alone is not a cryptographic proof — the runtime still has to present a key that matches the store.

Trust boundaries

A trust-store entry authorizes a named MCP server (and its keys) in that environment. It does not replace user authentication or FGA checks on user-owned resources. Deleting an entry removes the record; prefer revoke if you still need the audit trail.

Keys are environment-scoped. A rotated key does not revive a revoked entry.

Try it

Product context is on Agentic identity. The original release note is in the changelog.