ai-mcp

The MCP server for abap2UI5 — gives any AI coding agent (Claude Code, Cursor, VS Code Copilot, or any MCP client) the full abap2UI5 development loop, without an SAP system:

capabilities -> validate_view -> deploy_app -> build_backend -> run_app
 (what can        (static gates:     (write ABAP,    (transpile       (boot headless,
  I express?)      props + render)    lint)           to Node)         errors + SCREENSHOT)

The agent writes an ABAP class, validates the view in seconds, deploys it, boots it in a real browser and looks at the screenshot — then iterates. Everything runs locally on infrastructure that already guards the abap2UI5 ecosystem in CI: the abaplint transpiler + open-abap runtime, the framework's express shim, the ai-demokit build and boot gates, and the ai-view-check validation core.

Setup

The server orchestrates sibling checkouts (override locations with env vars):

git clone https://github.com/abap2UI5/abap2UI5        # A2UI5_HOME
git clone https://github.com/abap2UI5/ai-demokit      # AI_DEMOKIT_HOME
git clone https://github.com/abap2UI5/ai-view-check   # AI_VIEW_CHECK_HOME (optional, for validate_view)
git clone https://github.com/abap2UI5/ai-mcp
cd abap2UI5 && npm ci && cd ../ai-demokit && npm ci && cd ../ai-view-check && npm ci && cd ../ai-mcp && npm ci
npx playwright install chromium

Register in your MCP client, e.g. Claude Code:

claude mcp add abap2ui5 -- node /path/to/ai-mcp/server.mjs

Related MCP server: dassian-adt

Tools

run_app works for new dev apps and equally for the 276 existing ai-demokit ports and z2ui5_cl_ai_app_overview — useful as a reference: "run the closest existing port, look at it, then build mine".

The intended agent loop

1. capabilities { query: ... } — check the feature is expressible (and how) before writing a line of ABAP.

2. generation_rules — once per session.

3. Write the class, validate_view — fix findings in seconds.

4. deploy_app — abaplint against the full framework context.

5. build_backend — incremental after the first full build.

6. run_app — read the errors, look at the screenshot. Edit, validate, deploy, build, run again.

Notes

• Dev sandbox: deployed apps land in the ai-demokit checkout's gitignored src/zz_dev/ — nothing an agent deploys can leak into a commit. Promote a finished app by moving it into a real package deliberately.

• Port: the backend listens on 3000 (A2UI5_MCP_PORT overrides).

• UI5 sources: modules are served from the ai-demokit checkout's @openui5 packages, so booting needs no network. The built theme CSS is not in those packages — with network access it loads from the CDN (styled screenshots); without, apps render unstyled but structurally complete. A2UI5_MCP_OFFLINE=1 forces the hermetic behaviour.

• Chromium: uses the Playwright-managed browser; if absent, falls back to a system chromium (A2UI5_MCP_CHROMIUM overrides the executable path).

• Real system deployment stays what it is today: abapGit. This server is the inner dev loop; a run_app_system backend (launch URL + auth proxy, as solved in the VS Code extension) is the planned second stage.