SAP ABAP MCP
S/4HANA development inside an AI conversation
A local MCP server that exposes SAP ADT search, source reading, syntax diagnostics and controlled deployment to AI coding assistants — with design, dev and test subagents guarding Clean Core boundaries.
The problem
Traditional ABAP work bounces between SAP GUI, Eclipse ADT and documents. AI rarely sees real system context, and “change then verify” is hard to close safely.
Approach
- Wrap ADT capabilities in MCP: 17 high-frequency tools by default, with the rest discovered on demand through capability search to keep context small.
- Design, dev and test subagents have disjoint write scopes, preventing cross-boundary edits by process design.
- Extend along Clean Core tiers 1/2/3, keep writes limited to controlled objects and $TMP, and deploy only after remote syntax diagnostics.
What is inside
- 17 default tools plus an adaptive tool surface for system checks, code search, structure inspection and source diagnostics.
- Three subagents with their own definitions, JSON contracts and write boundaries, validated by script.
- Where-used impact analysis, DDIC reads, syntax diagnostics and ABAP Unit authoring support.
- Connection doctor, MCP smoke test and Codex integration config included, reproducible step by step.
Deploy it yourself
Prerequisites
Node.js 20+, a reachable SAP S/4HANA ADT endpoint, and an account with ADT permissions.
- Clone the repo, run npm install, and follow the README to configure the SAP connection and local MCP client.
- Run npm run mcp:doctor for a connection check, then npm run mcp:smoke to verify the protocol and tool surface.
- Add the MCP instance to Codex/Claude/Cursor, start read-only, and enable controlled writes only after verification.
Validate in a development client first and use a least-privilege account. Never put production passwords in the repo or prompts.
A system to ship, or a workflow to improve?
Tell me the current state, the outcome you need and your timeline. I reply within 24–48 hours and start with whether it is worth doing — and the fastest path.