maki-mcp

# maki Run a dev command, save its terminal output per project, and let MCP-capable agents read the logs without copy/paste. ```bash maki --name dev pnpm dev ``` Then ask your agent: > Check the maki dev logs and fix the error. ## How it works ```text maki --name dev pnpm dev ├─ runs pnpm dev ├─ prints output to your terminal └─ writes logs to .maki/runs/<run-id>.log maki-mcp --project /path/to/project └─ exposes MCP tools that read .maki logs ``` Each run gets: ```text .maki/ active.json runs/ 20260710-090000-a1b2c3.log 20260710-090000-a1b2c3.json ``` The run id is timestamp + random suffix. `--name` is the human label agents can use, like `dev`, `api`, `web`, or `worker`. ## Install Recommended: ```bash pipx install maki-mcp ``` or: ```bash uv tool install maki-mcp ``` On Windows without pipx: ```powershell py -m pip install --user pipx py -m pipx ensurepath pipx install maki-mcp ``` Avoid system-wide `pip install maki-mcp`. Some Python installs, like Homebrew Python, block it with an "externally managed environment" error. `pipx`/`uv tool` install maki as a CLI without touching system Python. Maki works in any project, not just Python projects: ```bash cd my-node-app maki --name dev pnpm dev ``` The package is `maki-mcp`, but the commands are: ```bash maki maki-mcp ``` ## Usage Start your app through maki: ```bash maki --name dev pnpm dev maki --name api npm run api maki --name worker python worker.py ``` If you omit `--name`, it defaults to `dev`: ```bash maki pnpm dev ``` Start the MCP server for that project: ```bash maki-mcp --project /path/to/project ``` MCP tools: - `list_runs()` - `latest_logs(name="dev", lines=200)` - `get_log(run_id="...", lines=200)` ## MCP config Use stdio transport. No port needed. ### Claude Code CLI: ```bash claude mcp add maki -- maki-mcp --project /path/to/project ``` Config shape: ```json { "mcpServers": { "maki": { "command": "maki-mcp", "args": ["--pro…