MCP

The MCP settings page lets external AI applications connect to your documentation through a hosted Model Context Protocol (MCP) server. Every Doccupine site exposes an MCP endpoint at /api/mcp that AI tools can query to search and read your content.

Connect your site with AI apps

The Connect your site with AI apps card gives you a ready-to-paste MCP configuration for popular AI tools. Pick a tab, copy the snippet, and add it to that tool's MCP config:

Claude - add the configuration to Claude Desktop's claude_desktop_config.json, or add the server URL as a connector from Claude's settings.
Cursor - add the configuration to Cursor's ~/.cursor/mcp.json, or connect it from Cursor's MCP settings.
Others - a generic mcpServers block for any MCP-compatible application.

Every snippet points to your site's own MCP endpoint, for example:

{
  "mcpServers": {
    "Your Project": { "url": "https://your-site.com/api/mcp" }
  }
}

The connection snippet appears once your site has been deployed. If you haven't published yet, deploy your site first to get its MCP server URL.

MCP server authentication

On a public site, the MCP endpoint is publicly accessible by default. When the site is password protected, MCP requires the same session unless you configure its dedicated API key:

Enable Require API key for MCP access.
Enter an API key.
Save.

Once enabled, clients must include the key as a Bearer token in the Authorization header of every request to /api/mcp.

The API key is stored as an environment variable (DOCS_API_KEY) on your deployment, not in a JSON file. After saving, a redeploy is triggered automatically to apply the change.

For the full reference on how the MCP server works - available tools, rate limits, and endpoint details - see the Model Context Protocol documentation.