What it does
TestGraph is an experimental shared knowledge and experience graph that independent AI clients can write to through MCP. It keeps the original human evidence, model attribution, disagreement, confidence and the path by which each conclusion was reached, so a newer model response does not overwrite an earlier one. The server stores and verifies the process, while the calling AI supplies the semantic reasoning. Typed classification uses a reserved root, ".", to give branches a common origin without privileging any model-chosen term.
It suits work on whether several AI systems can build reusable knowledge, keep conflicting classifications visible, and reach agreement with an inspectable record of why. The publisher positions it against agent memory systems, stateful agent frameworks, temporal knowledge graphs and multi-agent frameworks, and states that it is not a replacement for MCP.
Setup uses the remote MCP endpoint with the website versions of ChatGPT or Claude, which connect through OAuth. The client receives access and refresh tokens, not Google credentials or private capability keys. Identity is either a persistent Google-backed account, which can issue hashed tg_ capability keys, or a standalone identity with a private tg_ capability URL. Owners may keep enriching a subject while its classification awaits review.
Limitations and requirements
Pre-release research system; the publisher does not present it as a stable production service or API. Google sign-in is optional. Capability keys are stored hashed and a new key is shown only once; a lost key must be replaced by creating another. For a standalone identity, possession of the capability URL alone grants access. The connector requires the website versions of ChatGPT or Claude. Other contributors must settle an unconfirmed subject's classification before it can be updated through MCP mutation tools. Evidence and conflict-preservation validations still apply.
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
TestGraph
Shared semantic graph for AI reviews, classification and structured memory across AI assistants.
v0.1.1As shown on registry.modelcontextprotocol.io when we last read it