mind code
Browse documentation

Mind as an MCP server

Mind indexes your code, records what its reviews decided and keeps the conventions your team has agreed. Other coding agents cannot see any of that — they read the files in front of them and start from nothing each time.

mind mcp serve exposes those three things over the Model Context Protocol, so whichever agent you are working in can ask Mind instead of rediscovering them.

Connect your agents

One command, which finds the agents on your machine and asks which to connect:

mind setup

It adds an entry named mind to each agent you pick, pointing it at mind mcp serve. Restart the agent afterwards so it picks the server up.

mind setup --list          # what Mind supports, and what it found
mind setup --agent gemini  # connect one, without prompting
mind setup --yes           # connect everything detected
mind setup --remove        # undo it

The tools

search_code

Searches the code you have indexed by meaning rather than by string, and returns the matching chunks with their repository, file and line range.

list_findings

What review has already found in a repository, and what was decided — fixed, dismissed, or still open. An agent can check this before repeating a suggestion your team has ruled on.

list_memory

The conventions your organisation's agents have learned, with where each came from.

Every call is made as you, so an agent sees exactly the repositories your team visibility already allows — no more than the dashboard would show you.

Where your key lives

In ~/.mind/config.json, and nowhere else. The server reads it itself, so the other agent's configuration only ever names the command to run. Setup never copies your key into another program's files, and you can check that by reading any file it writes.

Mind writes that file so only your user can read it, and mind doctor warns if it ever becomes readable by anyone else.

Supported agents

Setup writes the configuration for these:

  • Claude Code — through claude mcp add, its own documented interface.
  • Gemini CLI — ~/.gemini/settings.json.
  • GitHub Copilot (VS Code) — ~/.copilot/mcp-config.json.
  • Antigravity — ~/.gemini/config/mcp_config.json.

For Augment and Continue, setup prints the steps rather than editing anything. Augment configures MCP through its own settings panel, and Continue uses a per-server file of its own — in both cases Mind would be guessing at a format, and it would be guessing about a file you depend on.

What setup changes

As little as it can. These files are yours, so setup:

  • keeps every other key and every other server;
  • copies the file to <file>.mind-backup before its first change;
  • leaves the file's permissions as they were;
  • refuses a file it cannot parse, names it, and changes nothing;
  • writes to a temporary file and renames it, so an interrupted run cannot leave half a config behind.

Undoing it

mind setup --remove

That takes out the mind entry and leaves everything else alone. The .mind-backup file beside each config is still there if you want the original back.

Troubleshooting

The agent says the server failed to connect. It runs mind from your PATH, so check that which mind finds it — an agent launched from a desktop icon may not have the same PATH as your shell.

The tools return an authentication error. Run mind whoami to check the stored key, and mind login if it has expired. The server refuses to start without one.

A repository is missing from results. The server shows only what your team visibility allows, and only what has been indexed. Check the repository on the dashboard first.