MCP Server

erpl-adt doubles as a Model Context Protocol server. Every CLI command is exposed as an MCP tool over JSON-RPC 2.0 on stdin/stdout. Plug it into any MCP-compatible agent — Claude Code, Cursor, Gemini CLI — and the agent decides which tools to call and in what order.

Start the server directly:

erpl-adt mcp --host sap.example.com --port 44300 --https

The process talks JSON-RPC on stdio, so you don't run this yourself; the MCP client launches it for you.

Client configuration

Claude Code

Add an entry to your ~/.claude/mcp.json (or per-project .claude/mcp.json):

{
  "mcpServers": {
    "sap": {
      "command": "erpl-adt",
      "args": ["mcp", "--host", "sap.example.com", "--port", "44300", "--https"],
      "env": { "SAP_PASSWORD": "..." }
    }
  }
}

Restart Claude Code and the sap tools appear in the tool list.

Cursor

Settings → MCP → + Add new MCP server and paste the same shape as above.

Gemini CLI

Add to ~/.config/gemini/config.yaml under mcp_servers::

mcp_servers:
  sap:
    command: erpl-adt
    args: ["mcp", "--host", "sap.example.com", "--port", "44300", "--https"]
    env:
      SAP_PASSWORD: "${SAP_PASSWORD}"

Tools the agent gets

The MCP tool names are derived from the CLI command paths. The most-used ones:

MCP tool	CLI equivalent	What it does
`adt_search`	`search`	Search ABAP objects by pattern and type
`adt_object_read`	`object read`	Read an object's metadata (URI, includes, source URIs)
`adt_source_read`	`source read`	Read source for a class, program, or include
`adt_source_write`	`source write`	Write source (auto-lock, transport, optional activate)
`adt_activate`	`activate`	Activate an inactive object
`adt_test`	`test`	Run ABAP Unit tests
`adt_check`	`check`	Run ATC quality checks
`adt_transport_create` / `_list` / `_release`	`transport …`	Transport lifecycle
`adt_ddic_table`	`ddic table`	Inspect a transparent table with check tables and abap types resolved
`adt_ddic_cds`	`ddic cds`	Read a CDS view's source
`adt_package_list` / `_tree` / `_exists`	`package …`	Browse package contents
`bw_search`, `bw_read_adso`, `bw_lineage`, `bw_export_cube`, …	`bw …`	Full BW/4HANA toolkit (see BW page)

46 tools are registered in total. Run erpl-adt mcp --list-tools to print the current set (returns a JSON manifest), or call the standard MCP tools/list method from your client.

Example agent loop

A typical session inside Claude Code looks like this — three CLI commands the agent picks itself, from one natural-language question.

You: What flight-related classes exist in this SAP system?

Claude calls adt_search with pattern ZCL_FLIGHT*:

Found 4 matching objects:

  Name              Type      Package       Description
  ────────────────  ────────  ────────────  ──────────────────────────
  ZCL_FLIGHT_CTRL   CLAS/OC   ZFLIGHT_APP   Flight booking controller
  ZCL_FLIGHT_MODEL  CLAS/OC   ZFLIGHT_APP   Flight data model
  ZCL_FLIGHT_TEST   CLAS/OC   ZFLIGHT_APP   Flight module unit tests
  ZCL_FLIGHT_API    CLAS/OC   ZFLIGHT_APP   REST API wrapper

You: Show me the booking controller and tell me what's broken.

Claude calls adt_object_read → adt_source_read → adt_test and writes back a diagnosis. The end-to-end narrative is in Your AI Coding Agent Just Learned ABAP and the follow-ups Real-Time SAP for AI, Part 1 and Part 2.

Authentication

Credentials are resolved in this order: CLI flags → --password-env env var → ~/.adt.creds (saved by erpl-adt login) → the SAP_PASSWORD environment variable.

For MCP usage, the env block in the client config is the cleanest path — the agent never sees the password, the password never appears in shell history, and you can rotate it without touching the client config.

Where to next

BW guide — the BW modeling tools available over MCP
Command Reference — every operation, also reachable via MCP
Announcement post — full context, demo recording, and motivation

Client configuration​

Claude Code​

Cursor​

Gemini CLI​

Tools the agent gets​

Example agent loop​

Authentication​

Where to next​