Use this guide when you want to install and configure the Autonomi MCP server against a running antd instance. It is the deeper setup guide for AI clients that talk to Autonomi through MCP tools instead of raw HTTP requests.
antd running on your machine or reachable at a known base URL (see Start the Local Daemon)
Python 3.10+
A local checkout of ant-sdk
antd Python SDK installed with REST support
From the ant-sdk repo root:
pip install "antd[rest]"
pip install -e antd-mcp/The antd-mcp package installs a command named antd-mcp.
The server still attempts daemon.port discovery before falling back to the default REST URL.
antd writes ant/sdk/daemon.port, while antd-mcp reads ant/daemon.port. Treat ANTD_BASE_URL as the reliable setup path unless you have already confirmed port-file discovery works in your environment.
If no environment variable or readable port file is available, it falls back to:
For Claude Desktop and other stdio-based clients:
For web-based clients that want SSE transport:
Set ANTD_BASE_URL explicitly for a reliable connection:
Add this to claude_desktop_config.json:
Adjust ANTD_BASE_URL if your daemon runs on a different host or port.
Your MCP client is configured correctly when it can see the antd-autonomi server and call a basic tool such as check_health() successfully.
The server cannot find the daemon: Confirm that antd is running, then set ANTD_BASE_URL explicitly.
Import or package errors: Make sure both antd[rest] and antd-mcp were installed into the same Python environment.
Tool calls fail immediately: Check the daemon first with curl http://localhost:8082/health.
Last updated
http://127.0.0.1:8082antd-mcpantd-mcp --sseANTD_BASE_URL="http://your-host:8082" antd-mcp{
"mcpServers": {
"antd-autonomi": {
"command": "antd-mcp",
"env": {
"ANTD_BASE_URL": "http://127.0.0.1:8082"
}
}
}
}