cap -plugin

This CAP MCP Plugin automatically converts annotated CAP services into an MCP server accessible to MCP clients. After installing the plugin and annotating your CAP services with @mcp and related tools, start your CAP application (typically via cds serve). The MCP server will be exposed under the /mcp endpoint on your running CAP app (e.g., http://localhost:4004/mcp). Use an MCP client or inspector to browse resources, tools, and prompts generated from your CAP data models. Tools created from annotated functions and actions will appear as executable MCP tools with descriptions and field hints to guide AI usage. You can also opt to expose CAP entities as resources with OData v4 query support, and wrap entities as tools for AI workflows. To experiment, connect the MCP inspector to http://localhost:4004/mcp to explore available resources and tools, then invoke tools or issue natural language queries against your data.

Prerequisites:

• Node.js 18+ installed
• SAP CAP tooling available (cds CLI)
• npm installed

Step 1: Install the MCP plugin

npm install @gavdi/cap-mcp

Step 2: Add MCP configuration to your CAP project (package.json)

{
  "cds": {
    "mcp": {
      "name": "my-bookshop-mcp",
      "auth": "inherit",
      "wrap_entities_to_actions": false,
      "wrap_entity_modes": ["query", "get"],
      "instructions": "MCP server instructions for agents"
    }
  }
}

Step 3: Annotate your CAP services with @mcp and related options as shown in the README examples.

Step 4: Start your CAP application (the MCP server runs as part of the CAP app)

cds serve

Step 5: Access the MCP endpoint at

• MCP Endpoint: http://localhost:4004/mcp
• Health Check: http://localhost:4004/mcp/health

Optional: Test with the MCP Inspector

npx @modelcontextprotocol/inspector

Note: The server is tied to the CAP app process. If you run in a container or with a process manager, ensure PORT and MCP_ENDPOINT align with your deployment configuration.

Tips and common issues:

• Ensure your CAP app exposes the MCP route at /mcp and is reachable from your client (adjust PORT if you run on non-default ports).
• When wrapping entities as tools, consider enabling mcp.wrap in CAP annotations to automatically generate executable tools like CatalogService_Books_query or CatalogService_Books_get.
• Use @mcp.omit to protect sensitive fields from MCP responses.
• If you modify annotations, restart the CAP server to regenerate MCP endpoints.
• The inspector can be used to explore resources, tools, and prompts and can help validate the generated MCP configuration.
• Environment variables like CAP_APP_BASE_URL, MCP_ENDPOINT, and PORT can be overridden in your deployment environment to suit containerized or multi-service setups.