MCP Integration

Kommit provides a full MCP server that AI coding tools can connect to. Your canvas becomes a queryable knowledge base — agents read your spec, features, tech stack, and project memory in real time.

What is MCP?

The Model Context Protocol (MCP) is an open standard for connecting AI models to external data sources and tools. An AI coding tool that supports MCP can pull structured context from any MCP-compatible source — including Kommit.

Quick Start

The fastest way to connect is with the Kommit CLI:

npx @kommit/cli@latest --client claude-code

This opens your browser to authenticate, then writes the MCP config to your project. Restart your editor and Kommit is connected.

The CLI supports 18 AI tools including Claude Code, Cursor, VS Code, Claude Desktop, Windsurf, Cline, Zed, and more. Run without --client for an interactive picker.

Manual Setup

If you prefer to configure manually:

1. Generate an API key

1. Go to Settings in your Kommit dashboard.
2. Under API Keys, click Generate New Key.
3. Copy the key (it starts with km_). You will only see the full key once.

2. Connect your AI tool

Claude Code — add to your project's .mcp.json:

{
  "mcpServers": {
    "kommit": {
      "type": "http",
      "url": "https://getkommit.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer km_YOUR_API_KEY"
      }
    }
  }
}

Cursor — add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "kommit": {
      "url": "https://getkommit.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer km_YOUR_API_KEY"
      }
    }
  }
}

VS Code — add to .vscode/mcp.json:

{
  "mcpServers": {
    "kommit": {
      "type": "url",
      "url": "https://getkommit.ai/api/mcp",
      "headers": {
        "Authorization": "Bearer km_YOUR_API_KEY"
      }
    }
  }
}

Other tools (Claude Desktop, Windsurf, etc.) use stdio transport via mcp-remote. The CLI handles this automatically — run npx @kommit/cli@latest and select your tool.

Available Tools

Once connected, your AI tool has access to 11 tools:

Spec & Project Tools

Memory Tools

Utility

How Agents Use It

A well-connected agent follows this workflow:

1. Load context — get_spec or get_project_overview to understand the project
2. Check memory — search_memory before making architectural choices
3. Get details — get_feature or get_tech_stack for specific requirements
4. Build — implement with full context
5. Save learnings — add_memory to record decisions and discoveries

Memories persist across sessions. The next time you (or a teammate) open a project, those decisions are part of the context.

Cross-Project Search

Memory tools work across all projects by default. Omit the project_id parameter to search or list memories across your entire workspace. Results include the project name for disambiguation.

To scope to a specific project, pass project_id or add ?project=PROJECT_ID to the MCP URL.

API Key Security

• Keys are scoped to your workspace. They cannot access other workspaces.
• Optional project scoping narrows access to specific projects.
• Keys can be revoked at any time from Settings.
• All requests are authenticated and subject to rate limits.

Why MCP Instead of Copy-Paste?

Copy-pasting Markdown works but has limitations:

• The spec is static — it does not update as implementation decisions are made
• The agent cannot query a specific section without reading everything
• When you regenerate the spec, the agent has no way to know

MCP solves all three: the agent reads the latest spec at query time, requests only what it needs, and always gets the most recent version.