What it does
It scans local Claude Code transcripts and Cursor sessions for three kinds of signal: instructions repeated across sessions, corrections made after the agent went wrong, and rejected tool calls. From these it distills proposed rules, each with the user's own quotes as evidence and a confidence rating. Accepted rules are written into a managed block in both CLAUDE.md and AGENTS.md, and hand-written content is left untouched.
It suits teams and individuals who keep re-teaching the same conventions to coding agents, or whose context files have fallen out of date. A global mode mines all projects for personal working preferences and maintains a managed block in ~/.claude/CLAUDE.md. A promote command reviews saved agent memory for rules that belong globally, and a stale command flags context-file references that no longer match the repo, exiting with code 1 on findings so it can run in CI.
The tool is a command-line interface run from a terminal, with a Claude Code plugin that adds a session-start check for new signals, and an MCP server that runs the approval loop inside chat. Distillation uses the existing claude CLI, so no separate API key is needed. Proposals are applied only after an explicit decision, and rejected titles are not proposed again.
Limitations and requirements
Distillation requires the local claude CLI to be installed. Cursor session mining uses node:sqlite and needs Node 22 or later; on older Node versions the Cursor source is skipped. Scanning reads local session folders such as ~/.claude/projects. No context file is changed without the user's approval.
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
io.github.ChiragBachani/context-autopilot
Mines your Claude Code and Cursor sessions into evidence-based CLAUDE.md / AGENTS.md rules.
v0.6.0As shown on registry.modelcontextprotocol.io when we last read it