@cap-js/mcp-server

What it does

This server exposes SAP CAP project structure to AI models through two specialized tools. search_model performs fuzzy search over compiled CDS (Core Data Services) definitions—entities, fields, services, and HTTP endpoints extracted from .cds files. search_docs provides semantic search through CAP documentation using vector embeddings, allowing queries phrased naturally without exact keyword matching. Together, they enable AI models to ground code generation in verified project schemas and official CAP patterns.

Who it's for

SAP CAP developers using AI-assisted editors (VS Code extensions like Cline, GitHub Copilot, Claude Code, or CLI environments) to accelerate cloud application development. Teams building enterprise applications on SAP Cloud Application Programming Model who want their AI coding assistants to query live project schemas and documentation.

Common use cases

• Query CDS entities and their relationships to understand data models and HTTP service endpoints
• Search CAP documentation for specific patterns (e.g., "how to add columns to a select statement in Node.js") during development
• Ground AI-generated code suggestions with verified CAP examples and entity schemas
• Enable agentic workflows where AI assistants self-serve documentation and schema queries without blocking developers

Setup pitfalls

• The server reads your project's .cds files; ensure the MCP client runs with the CAP project as its working directory or can access the project path
• Vector embeddings for documentation search are bundled in the npm package; offline or air-gapped environments must download the full package before use
• Configuration differs by client: VS Code requires mcp.json or extension-specific config; verify your client's MCP setup documentation
• The tool_count registry metric incorrectly shows 0; the server actually provides search_model and search_docs tools