Authdog
Back to journal
Product UpdatesMCPAgents

Register AI agents and machine identities

A console Agents module lets teams register scoped keys, pin tool manifests, and audit the machines that call APIs and MCP servers.

From the changelog: Agent and machine identity management

Authdog Team

2 min read
Centered Authdog and Model Context Protocol marks on a grainy teal wash with dashed orbits and satellites

Human users are not the only callers of your APIs. Agents and other machines need an identity you can register, scope, rotate, and revoke. Authdog now exposes that registry as a first-class Agents console module, built on the MCP trust store.

This is the longer note behind the August 25 changelog entry.

What an agent identity is

Each entry is an environment-scoped trust record: a name, optional subject, verification status (unverified, verified, revoked), allowed scopes, allowed tools, and one or more keys (fingerprint, algorithm, expiry, revocation).

You can pin a tool manifest (and its hash) so the tools an agent is allowed to call are reviewable and do not silently expand. Delegation and key rotation stay on the same record; revoke a key or the whole entry when the workload should stop.

Activity lands in a dedicated, filterable audit view so agent calls are not mixed into generic user sign-in noise. MCP runtime events (MCP_TOOL_INVOKED, MCP_TOOL_DENIED, MCP_AUTH_DENIED, MCP_TRUST_VIOLATION) are the telemetry the SDK emits on the wire; the Agents module is the control-plane registry those calls are checked against.

Set it up

In the console, open Agents:

  1. Register an identity and set allowed scopes and tools.
  2. Attach a key and record the fingerprint your runtime will present.
  3. Optionally pin a tool manifest so allowed tools cannot drift.
  4. Mark the entry verified when you have reviewed it; revoke it to cut access.

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

Trust boundaries

An agent identity authorizes machine access to tools and APIs you listed. It does not replace user authentication or FGA checks on user-owned resources. A pinned manifest is only as good as the hash you stored; rotate keys when a workload is rebuilt or a secret leaks.

MCP management (server CRUD in the console) and MCP runtime (tool-call telemetry) are different categories. Do not conflate a registered agent with a published MCP server.

Try it

Open Agents in an environment that uses MCP. The original release note is in the changelog.

More from Authdog
ShareLinkedInX