# Armature

Session analytics and evals for MCP servers, Claude connectors and ChatGPT apps

- Website: https://armature.tech/usability
- Listing: https://openack.co/p/armature
- Not evaluated. Listed by OpenACK: its 10 claimed capabilities are OpenACK's reading of the product's public site, not the maker's word, and nobody has checked them yet.

OpenACK's description (not checked):

> Armature captures the sessions AI agents run against your MCP server, rebuilds each one with the user's intent and every tool call, and groups them into use cases and issues. It also runs eval suites that replay user goals against the server and score the result.

## Capabilities

- **not evaluated**: Register an MCP server in a workspace
- **not evaluated**: Create an ingest API key for an MCP server
- **not evaluated**: Replay an agent session with every tool call
- **not evaluated**: Filter agent sessions by outcome
- **not evaluated**: See sessions grouped into use cases ranked by volume and pass rate
- **not evaluated**: Archive an issue in the issues inbox
- **not evaluated**: Create an eval case and run it against your MCP server
- **not evaluated**: Group eval cases into a suite and run the suite
- **not evaluated**: Invite a teammate to a workspace
- **not evaluated**: Connect an AI assistant over MCP

## What the statuses mean

- **confirmed** (ACK): a person from OpenACK signed in with a test account and saw it work.
- **not confirmed** (NAK or ?): it was tried and didn't work, or we couldn't reach or check it. Not proof it's missing, but not verified.
- **not evaluated** (—): nobody has checked it. It's the founder's word only.

Treat only "confirmed" as verified. A result is a record of one evaluation on the date shown, not a guarantee.