# MCP Server Mode (`grokkit mcp`)

Grokkit can run as an **MCP (Model Context Protocol)** server, allowing AI coding agents like **Claude Code**, Cursor, or other MCP-compatible clients to call Grokkit tools directly during their workflows.

## What is MCP?

MCP is a protocol that lets LLMs call external tools and access resources in a standardized way. By running `grokkit mcp`, you turn Grokkit into a service that other AI tools can use.

## Starting the MCP Server

```bash
grokkit mcp
```

This starts the server over **stdio** (standard input/output), which is the default transport used by Claude Code and most MCP clients.

The server will run until the client disconnects.

## Available Tools

When running in MCP mode, the following tools are available:

- **`lint_code`** — Run linter on a file and return results
- **`analyze_code`** — Deep project analysis with educational report
- **`generate_docs`** — Generate documentation comments for a source file
- **`generate_tests`** — Generate unit tests for supported languages
- **`generate_commit_msg`** — Generate conventional commit message from staged changes
- **`run_recipe`** — Execute a named recipe from `.grokkit/recipes/`

## Configuration

You can control MCP behavior in your config file (`~/.config/grokkit/config.toml` or `./config.toml`):

```toml
[mcp]
enabled = true
tools = ["lint_code", "analyze_code", "generate_docs"]
resources = ["recipes://local", "prompts://language"]
```

## Example Usage with Claude Code

Once `grokkit mcp` is running, you can tell Claude Code:

> "Use the grokkit MCP server to lint this file and then generate tests for it."

Claude can then call the tools directly.

## Resources

MCP also exposes resources:

- `recipes://local` — List of local recipes
- `recipes://global` — List of global recipes
- `prompts://language` — Available language prompts

---

**See also:**
- [Developer Guide](../developer-guide/mcp.md) for technical details
- [Main README](../../README.md) for installation