What it does
Metis captures tacit fragments of expert practice: the cues an expert noticed, how they responded, and the circumstances of that response, which procedures and logs usually leave out. After human review, fragments sit alongside procedures, facts and past events in an agent's memory. Each fragment is tagged with one of seventeen categories of tacit knowledge (K1 to K17). Every capture, confirmation, review decision and retrieval is recorded on a hash-linked evidence chain through the CHAP reference coordinator.
It suits teams that want agents to draw on expert judgement as well as documented procedures, and that need a record of how that judgement was reviewed and used. The MCP server exposes the same memory to clients such as Claude Desktop and Claude Code, and can run without prior installation. A demo works from supplied observations without a model server. Connectors capture from workplace systems and send short questions to the expert ("whispers") in Slack or Teams.
Local use keeps records in a local folder. For team use, the documentation covers sign-in, workspace roles, a web app and PostgreSQL, with deployment via Docker or Kubernetes. The application using Metis must supply its own capture tools, consent workflows, access control, reviewer identity, permissions and storage policy.
Limitations and requirements
Production use requires worker consultation, legal review and domain validation, and the publisher asks users to read its ethical use document first. Metis must not be used for covert monitoring. It records no audio, video, biometrics, screenshots or keystrokes. The integrating application must provide capture tools, consent workflows, access control, reviewer identity, formal change control, action limits and human escalation, and retention policy.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
Metis
Governed tacit memory for AI agents: reviewed expert know-how, given only where conditions match
v0.1.6As shown on registry.modelcontextprotocol.io when we last read it