Neusis Neuron
Neusis Neuron (neusis-neuron-mcp) is a minimal Model Context Protocol (MCP) server that gives the Neusis Code agent read-only access to a single Project Brain knowledge-base (KB) repository on GitHub.
The agent uses it to ground its work in your project's accumulated context — architecture decisions, the knowledge graph Project Brain generates, and ingested Jira / Confluence artifacts — without you pasting that context in by hand.
It is a fork of GitHub's official Go MCP server, locked down so the binary cannot be repurposed into a general GitHub agent: no issues, no pull requests, no writes, and no access to any repo other than the one it is bound to.
:::info Replaces Project Brain MCP Neusis Neuron supersedes the earlier Project Brain MCP server. If you used that one, see Migrating from Project Brain MCP. :::
How it works
You bind the server to one KB repository at startup with a required --kb-repo OWNER/REPO flag. From then on:
- Every tool call operates only on the bound repo.
ownerandrepoare not parameters the agent can set — they are stripped from the tool schemas and injected by the server. The agent can never retarget the server at a different repository. - The tool surface is locked to three read-only tools — read a file, list the file tree, and search code — all confined to the bound repo. Every other GitHub MCP tool is masked and unreachable. See Tools.
- The server is a thin GitHub adapter. It holds no local state, keeps no cache, and never calls the Project Brain HTTP API — it talks only to the GitHub REST API.
Neusis Code ──stdio──▶ neusis-neuron-mcp ──▶ GitHub REST API ──▶ bound KB repo (read-only)
What you need
- Neusis Code, where you register the server — see Configuration.
- A GitHub knowledge-base repository managed by Project Brain (or any repo you want the agent to read).
- A read-only fine-grained GitHub token with
Contents: Readon that repo.