CAP MCP Server: an assistant for SAP CAP projects
CAP MCP Server
An SAP MCP server that searches the CDS model and CAP docs embeddings locally so an agent writes more accurate app code
- 118 ★ on the repository
- Last commit
- Apache-2.0
Minimal access
Why this level
- The tools search the project model and docs and write nothing
- Embedding search runs locally and project source is not sent out
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/cap-mcp-serverDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The tools search the project model and docs and write nothing
- Embedding search runs locally and project source is not sent out
This is third-party code. Review the repository files before installing.
What it does
@cap-js/mcp-server is an MCP server for the SAP Cloud Application Programming Model (CAP) built by the cap-js team. It helps an agent answer project questions: which CDS services are in it and where they are served, what the entities are about and how they relate, how to add columns to a select statement in CAP Node.js. There are two tools. `search_model` does fuzzy search over definition names in the compiled CDS model: the compiler merges all `.cds` files into one representation with definitions, relations, annotations and HTTP endpoints, and the search finds the entity Books even from the query "book". `search_docs` searches preprocessed CAP documentation through vector embeddings: the query becomes a vector and the closest fragments by meaning are returned. This runs locally with no external service. The server starts with `npx -y @cap-js/mcp-server` and suits Claude Code, Codex, opencode, Cline, GitHub Copilot and VS Code. The same tools are available in a terminal with `cds-mcp` after a global install of the package.
Who it is for. SAP CAP developers who write CDS services and models with Claude Code, Codex or Copilot and want the agent to know their own project and the current documentation.
Good fit when
- A project has many CDS services and the agent must find out which of them is served where
- You need to find an entity or definition by a partial name in a large model
- The agent writes a CAP Node.js query and gets the API wrong, so a docs page matched by meaning helps
- You onboard a new developer and ask the agent to explain entity relations
- You want the same searches from a terminal without an MCP client
Not a fit when
- The project is not on SAP CAP: the server only knows the CDS model and CAP documentation
- You need to write or change the model: the tools only search and write nothing
- You need stable interfaces for scripts: the author notes the tools are meant for AI models and do not form a stable API
Example request
Which CDS services are in this project and where are they served? Then show how to add a column to a select statement in CAP Node.js.MCP
- Transport
- stdio
- Authentication
- not required
Environment variables
CDS_MCP_MODEL- Embedding model for documentation search, for example nomic-ai/nomic-embed-text-v1.5.
Limitations
The tools are meant for AI models and do not count as a stable API: their names and parameters may change. The server searches what is in the project and in the bundled documentation and does not replace checking the official site cap.cloud.sap.
The embedding model can be changed with `CDS_MCP_MODEL`, for example to `nomic-ai/nomic-embed-text-v1.5`. If the package is installed globally, `npx` uses the installed version and does not download a fresh one, so update it by hand with `npm update -g @cap-js/mcp-server`.
CAP itself must be set up following the Getting Started guide on the SAP site. Security issues go through the repository's disclosure procedure.
How to disable. Remove the server with `claude mcp remove cds-mcp` or delete the `cds-mcp` entry from the client configuration. A global package is removed with `npm uninstall -g @cap-js/mcp-server`.
FAQ
How do I connect it to Claude Code?
Run `claude mcp add --scope user cds-mcp -- npx -y @cap-js/mcp-server` for all projects or `--scope project` to commit it to the current repository. An environment variable goes in with `--env` before the server name.
Does it work with Codex?
Yes, with `codex mcp add cds-mcp -- npx -y @cap-js/mcp-server`. The README has blocks for VS Code, Cline, opencode and GitHub Copilot.
What does search_docs search?
Preprocessed CAP documentation through vector embeddings: the query becomes a vector and the closest fragments by meaning are returned. The search runs locally on your machine and exact keywords from the docs are not required.
What does search_model do?
It fuzzy-searches definition names in the compiled CDS model: entities, services, annotations and endpoints. A partial name like "book" finds the entity Books.
Can I use it without an agent?
Yes. After `npm i -g @cap-js/mcp-server` the `cds-mcp` command appears, for example `cds-mcp search_docs "query" 1` or `cds-mcp search_model . Books entity`.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server for SAP documentation: unified search across SAPUI5, CAP, ABAP, BTP and RAP with BM25 full-text ranking.
SAP Fiori MCP Server: Fiori apps with AI
SAP Fiori MCP Server
SAP's official MCP server: an agent creates and edits SAP Fiori elements apps from a plain-language description
MCP server to SAP ABAP through ADT: an agent reads code of programs, classes, tables and CDS and searches objects in the system
ARC-1: an MCP server for SAP ABAP systems, safe by default, with reading, code edits, transports and checks through ADT.
add-mcp writes an MCP server into the configs of 25 agents with one command: Claude Code, Codex, Cursor, VS Code, Zed and more
An MCP server that lets an agent read files, search code and run commands in a workspace
README in short
A short description of the server, the questions it helps answer and setup for Claude Code, VS Code, Codex and opencode. It then describes both tools, command-line use, environment variables, the JavaScript API and how to reach support.