Skip to content

MCP Configuration

You can configure CiteKit to work with any MCP-compliant AI agent or IDE.

Claude Desktop

Add CiteKit to your claude_desktop_config.json:

json
{
  "mcpServers": {
    "citekit": {
      "command": "python",
      "args": ["-m", "citekit.cli", "serve"]
    }
  }
}

Cursor / Cline / Roo Code

Most modern IDE extensions support MCP via custom command execution.

  1. Open the MCP settings in your IDE.
  2. Add a new server named CiteKit.
  3. Command: python
  4. Args: ["-m", "citekit.cli", "serve"]

CLI Options for serve

The serve command supports the following environment variables and flags through the SDK and CLI:

Storage & Output

By default, CiteKit looks for maps in .resource_maps/ and saves output to .citekit_output/. You can change these by setting up a CiteKitClient configuration, though the serve command currently uses defaults.

Remote execution (Npx)

If you don't want to install CiteKit globally, you can run the JS MCP server via npx:

bash
npx -y citekit serve

Troubleshooting

  • Path Errors: Ensure python or node is in your system PATH.
  • Permission Errors: AI agents running MCP servers require permission to write to the outputDir. Ensure the directory is writable by the process.
  • Missing Maps: If listResources returns an empty array, run citekit ingest on some files first!

Released under the MIT License.