Developer Docs MCP

Benchling hosts a remote MCP server for these developer docs, so AI coding tools can search them directly. Point Cursor, Windsurf, Claude Code, or Claude Desktop at it and your assistant can answer Benchling questions from the current documentation instead of from memory.

The Developer Docs Model Context Protocol (MCP) server lets AI-powered editors such as Cursor and Windsurf, and general-purpose assistants such as Claude Code and Claude Desktop, search the Benchling developer documentation as part of a conversation. It only searches the docs. It is separate from the Benchling MCP, which connects AI tools to the Benchling agent.

What is MCP?

Model Context Protocol is an open standard for connecting AI applications to external data sources and tools. A client connects to an MCP server, discovers the tools it offers, and calls them on the user’s behalf.

The server exposes documentation search. Your assistant can look up how an endpoint behaves, what a manifest field means, or which schema a warehouse table maps to, and answer from the current docs rather than from whatever it happened to memorize. That matters most for the parts of the platform that move quickly, especially the V3 REST API surface.

No credentials required

Documentation search reads public content, so the server needs no API key and no OAuth token. Connecting it does not give an AI tool access to your tenant’s data.

Connecting a client

Benchling hosts the server at https://docs.benchling.com/mcp. Add it to your client’s MCP configuration:

{
  "mcpServers": {
    "benchling": {
      "url": "https://docs.benchling.com/mcp"
    }
  }
}

Restart the client after editing its configuration file so it picks up the new server.

If your client requires custom headers for remote servers, it can set them the way it sets them for any other MCP server; the Developer Docs MCP does not require any.

Checking the connection

Once the client has connected, confirm it can reach the server:

Open your editor or assistant and start a new chat.
Ask a question that can only be answered from the documentation, something specific enough that a general model would not know it, such as how V3 pagination tokens work or which stability tier a given endpoint sits in.
Look for the client indicating that it called the server’s searchDocs tool, rather than answering directly.

If the assistant answers without invoking the server, it is answering from memory. Most clients list their connected MCP servers and the tools each one exposes; check there to confirm the server registered.