Connect via MCP

The Mixlar docs run a live MCP server (Model Context Protocol) so your AI agent can search and read the SDK on demand — instead of loading the whole reference up front. Point a client at one URL and it gains two tools: search_docs and read_docs.

Endpoint

https://docs.mixlar.net/mcp.php

It’s a stateless Streamable HTTP MCP server — no install, no key. (For the whole reference as a single file instead, see Build with an AI agent.)

Connect it

Add to .cursor/mcp.json in your project (or the global one):

{
  "mcpServers": {
    "mixlar-docs": { "url": "https://docs.mixlar.net/mcp.php" }
  }
}

In a client that supports remote MCP / connectors, add a server with the URL:

https://docs.mixlar.net/mcp.php

If your client only speaks stdio, bridge to the remote server with mcp-remote:

{
  "mcpServers": {
    "mixlar-docs": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://docs.mixlar.net/mcp.php"]
    }
  }
}

The tools

Tool	Arguments	Returns
`search_docs`	`query` (string), `limit` (int, optional)	The most relevant documentation sections.
`read_docs`	`query` (string, optional)	The full text of the best-matching section — omit `query` for the overview.

Try it

Once connected, ask your agent something like:

Using the mixlar-docs tools, look up how device widgets push data, then write a plugin with a widget that shows my CPU temperature. Scaffold it with mixlar-sdk create and validate it.