klypix-mcp: shared memory for Claude Code, Codex and Cursor

Install in your project root

Open a terminal in your project folder, then run:

node --version
npx klypix-mcp install
npx klypix-mcp doctor

The installer creates a brain if needed and configures detected coding tools. It writes project and home-directory settings; review the permissions. Reopen your coding tool or reconnect its MCP server, and allow the tools when your host asks.

Interactive installation enables on-device semantic search by default. Use --no-semantic to skip it, or --semantic for noninteractive installation. It needs an initial model download; without it, retrieval uses keywords. The Claude plugin path differs.

How does each coding tool connect?

See Cline, Copilot, Gemini CLI, Windsurf and Aider setup and integration limits.

Try a decision, a fresh session and a correction

Use a disposable trial project, or replace this example with a real decision before saving. These are MCP tool calls for your agent, not terminal commands. Replace /absolute/path/to/project with your absolute project path. files: [] means no source-code edits are planned.

1. Record the decision and reason

brain_sync { project: "/absolute/path/to/project", phase: "start",
  intent: "Record the trial storage decision", files: [] }
brain_note { area: "Storage",
  text: "Use SQLite for the trial project database because this prototype runs on one machine." }
brain_sync { project: "/absolute/path/to/project", phase: "complete" }

2. Open a fresh session in the same project

Optionally switch to another supported tool. Ask it to retrieve the decision:

brain_sync { project: "/absolute/path/to/project", phase: "start",
  intent: "Review the trial storage decision", files: [] }
brain_ask { question: "Which database does the trial project use, and why?" }

Check that the returned card names SQLite and the one-machine rationale.

3. Correct the decision and ask again

brain_note { area: "Storage",
  text: "CORRECTION: Use PostgreSQL instead of SQLite for the trial project database because the trial now needs concurrent writes from several machines." }
brain_ask { question: "Which database does the trial project use now, and why?" }
brain_sync { project: "/absolute/path/to/project", phase: "complete" }

Inspect the cards: PostgreSQL should be current and SQLite retained as superseded history. Name the earlier decision explicitly; CORRECTION alone does not guarantee a match. This checks your capture and recall path, not large-project retrieval quality.

Share the record with your team through Git

Commit brain.klypix and managed project configuration. Install in each clone, pull updates and review changes. Installation registers the card-level merge driver in Git repositories; an unconfigured clone gets an ordinary binary conflict. Follow the team setup. Keep stable rules and approved architecture records alongside the brain.

Limits and troubleshooting

• Tools missing? Reconnect the host and inspect doctor. A clean configuration does not prove the host loaded the tools.
• Decision missing? Check the project path and successful brain_note capture. A fresh brain cannot reconstruct unrecorded history.
• Standalone coordination is local. Live presence and file-overlap warnings stay on one machine and OS user. Warnings are advisory.
• Check the privacy boundary. Retrieval is local; context sent to a cloud agent can reach its provider. Model setup and updates can use the network. Review the permissions.
• Removal: the uninstall instructions distinguish machine installation from each project’s files. Do not hand-edit brain.klypix.