Overview
Rutter is a personal reference implementation for retaining decisions made during AI-assisted work. It stores dated, append-only records in a Markdown notes folder owned by the user, helping a later session recall what was decided and which notes were cited at the time.
The MCP server is designed for people who already keep Markdown notes and are comfortable configuring an AI client. It can be used with Claude Code, Grok, Codex, and Antigravity so the recorded thread can carry across supported tools.
Key Features
- Stores each captured decision as a line in a dated Markdown file.
- Records the path and content hash of every note cited by the client.
- Preserves earlier decisions through an append-only record rather than rewriting history.
- Resolves a renamed note when exactly one current note matches its stored hash.
- Displays unresolved references instead of guessing when a reliable match is unavailable.
How It Works
When an AI session makes a decision, the configured client writes a summary line with its note references. Rutter stores that line and the cited notes’ hashes as they existed at capture time. Later queries can return the decision, its date, and the associated references from the same AI tool or another configured client.
Reference resolution happens when a record is displayed, leaving the original stored line unchanged. If a note moved without changing, its hash can identify the new path; if it changed as well as moved, Rutter can mark the old reference unresolved.
Scope and Limitations
A record exists only when the client writes the required line, and a listed reference proves only that the session cited a note—not that the model read it. Changes made to a note in place at the same path are not currently flagged. Rutter is MIT-licensed, presented as one person’s tool and reference implementation, and carries no support commitment.