docs / 14

Using Kadhai from other apps (MCP and ACP)

As an MCP server

Other AI tools (Claude Desktop, Claude Code, Cursor) can use the crew and the folder through MCP. Tools offered: kadhai_run_task, workspace_list_files, workspace_read_file, workspace_write_file, workspace_delete_file, workspace_run_command, memory_read, locks_list, providers_status.

Over stdio, add to the tool's MCP settings:

{
  "mcpServers": {
    "kadhai": { "command": "node", "args": ["C:/path/to/Kadhai/bin/mcp-server.js"] }
  }
}

With Claude Code: claude mcp add kadhai -- node C:/path/to/Kadhai/bin/mcp-server.js. Add "--workspace", "C:/some/folder" to point it at another folder.

While npm start is running, the same tools are also served at http://localhost:4747/mcp, and tasks started that way appear live in the dashboard. The stdio server has no dashboard for approvals, so commands only run there with EXEC_MODE=auto.

As an ACP agent (editors)

Editors that speak ACP, such as Zed, can use the crew in their own agent panel, on the project folder they have open:

{
  "agent_servers": {
    "Kadhai": { "command": "node", "args": ["C:/path/to/Kadhai/bin/acp-agent.js"] }
  }
}

In this mode the workspace is your real project. Kadhai adds memory.md and .kadhai/ there (add both to .gitignore) and writes files straight to disk, so save open buffers first.

This page comes from the user guide in the repository. Something unclear or wrong? Tell us.