Your AI Coding Agent Just Learned ABAP

February 13, 2026 · 7 min read

Co-Founder & CEO

We built erpl-adt because we got tired of Eclipse being the only way to interact with ABAP systems programmatically. If you wanted to search objects, read source code, or run unit tests against SAP — you had to use Eclipse ADT. No CLI, no scriptable API client, nothing you could plug into a CI pipeline or hand to an AI agent.

So we built one. A single binary that speaks the ADT REST API, works from the command line, and doubles as an MCP server for AI coding agents.

erpl-adt CLI demo: search, read, test, fix, verify

Why This Exists

Eclipse ADT communicates with SAP over a well-structured REST API. The protocol itself is actually quite good — proper resource URIs, XML payloads, stateful sessions for locking. The problem was never the protocol. It was that Eclipse was the only client for it.

That means no VS Code extension that actually works against the real API. No way to script searches or automate source reads. And when the rest of the development world started using AI coding agents like Claude Code, Cursor, and Gemini — ABAP developers were locked out.

What erpl-adt Does

erpl-adt is a single, statically-linked C++ binary. No JVM, no SAP RFC SDK, no Eclipse. It runs on Linux, macOS, and Windows and works in two modes:

CLI — 20 commands across 9 groups. Search objects, read and write source, run unit tests, manage transports, browse data dictionaries.
MCP Server — the same operations exposed as Model Context Protocol tools. You can plug it into Claude Code, Gemini CLI, Cursor, or any MCP-compatible agent.

The MCP mode is the interesting part. It means any AI coding agent can talk to your SAP system — search code, read source, run tests, write fixes — through a standardized protocol. The agent decides which tools to call and in what order.

Demo: Claude Code Explores an ABAP System

Here's what a session looks like. Claude Code is connected to a SAP system via erpl-adt's MCP server — no scripting, just a normal conversation.

Setup — three lines in your Claude Code MCP config:

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

The conversation:

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 source of the booking controller.

Claude calls adt_read_object to get the structure, then adt_read_source on the main include:

CLASS zcl_flight_ctrl DEFINITION PUBLIC FINAL CREATE PUBLIC.
  PUBLIC SECTION.
    METHODS book_flight
      IMPORTING iv_carrid TYPE s_carr_id
                iv_connid TYPE s_conn_id
                iv_fldate TYPE s_date
      RETURNING VALUE(rv_booking_id) TYPE s_book_id
      RAISING   zcx_flight_error.

    METHODS cancel_booking
      IMPORTING iv_booking_id TYPE s_book_id
      RAISING   zcx_flight_error.
  ...
ENDCLASS.

You: Run the unit tests for this class.

Claude calls adt_run_tests:

ABAP Unit Results — ZCL_FLIGHT_CTRL

  Method                      Time      Result
  ──────────────────────────  ────────  ──────
  test_book_valid_flight      12ms      PASS
  test_book_invalid_carrier   8ms       PASS
  test_book_past_date         5ms       PASS
  test_cancel_existing        15ms      PASS
  test_cancel_nonexistent     6ms       PASS
  test_double_booking         9ms       PASS

  6/6 passed (55ms total)

You: What does the underlying SFLIGHT table look like?

Claude calls adt_read_table:

Table: SFLIGHT — Flight schedule

  Field     Type    Key   Description
  ────────  ──────  ────  ─────────────────
  MANDT     CLNT    *     Client
  CARRID    CHAR    *     Airline code
  CONNID    NUMC    *     Connection number
  FLDATE    DATS    *     Flight date
  PRICE     CURR          Airfare
  CURRENCY  CUKY          Currency key
  PLANETYPE CHAR          Aircraft type
  SEATSMAX  INT4          Max. capacity

Claude figured out which tools to call, in what order, and how to present the results. The agent has full access to the SAP system through erpl-adt's MCP tools.

Same System, Different Agent: Gemini CLI

Here's the same SAP system, this time accessed through Google's Gemini CLI. Same erpl-adt binary, same MCP config:

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

You: I need to understand the ZFLIGHT_APP package. What's in it?

Gemini calls adt_package_tree:

Package tree: ZFLIGHT_APP (12 objects)

  Type      Name                  Package          Description
  ────────  ────────────────────  ───────────────  ──────────────────────
  CLAS/OC   ZCL_FLIGHT_CTRL       ZFLIGHT_APP      Flight booking controller
  CLAS/OC   ZCL_FLIGHT_MODEL      ZFLIGHT_APP      Flight data model
  CLAS/OC   ZCL_FLIGHT_TEST       ZFLIGHT_APP      Flight module unit tests
  CLAS/OC   ZCL_FLIGHT_API        ZFLIGHT_APP      REST API wrapper
  PROG/P    ZFLIGHT_REPORT        ZFLIGHT_APP      Flight status report
  TABL/DT   ZFLIGHT_LOG           ZFLIGHT_APP      Booking log table
  DEVC/K    ZFLIGHT_SUB           ZFLIGHT_APP      Sub-package
  CLAS/OC   ZCL_FLIGHT_CLEANUP    ZFLIGHT_SUB      Data cleanup utility
  PROG/P    ZFLIGHT_MIGRATION     ZFLIGHT_SUB      Migration program

You: Run ATC checks on the whole package. Are there quality issues?

Gemini calls adt_run_atc:

ATC Results — ZFLIGHT_APP (variant: DEFAULT)

  2 warnings, 0 errors

  Priority  Object              Message
  ────────  ──────────────────  ──────────────────────────────────────
  Warning   ZCL_FLIGHT_MODEL    Method GET_FLIGHTS has cyclomatic complexity > 10
  Warning   ZFLIGHT_REPORT      Program contains unused variable LV_TEMP

You: Fix the unused variable in ZFLIGHT_REPORT.

Gemini calls adt_read_source, removes the unused lv_temp declaration, then calls adt_write_source — which handles the lock/write/unlock cycle automatically:

Source updated: /sap/bc/adt/programs/programs/zflight_report/source/main
Lock acquired, source written, lock released.

You: Verify — run syntax check and ATC again.

Syntax check: 0 errors, 0 warnings
ATC check:    1 warning (cyclomatic complexity), 0 errors

Unused variable warning resolved.

Two different AI agents, same SAP system, same MCP protocol, same results. That's the whole point — erpl-adt doesn't tie you to a specific AI vendor. It speaks a standard protocol, so any MCP-compatible agent works.

What erpl-adt Handles for You

The SAP ADT protocol has a few tricky parts that erpl-adt takes care of so your agent (or your script) doesn't have to:

CSRF token management — automatic fetch, cache, and refresh on 403
Stateful sessions — context IDs and cookies for lock/write/unlock workflows
Lock guards — automatic unlock on scope exit, even if the agent crashes mid-edit
Async polling — operations that return 202 Accepted (like activation) are polled to completion
Error categorization — structured error codes so the agent can decide whether to retry (auth failure vs. lock conflict vs. not found)

The Full Command Set

Group	Commands	What it does
`search`	`query`	Find ABAP objects with wildcards and type filters
`object`	`read` `create` `delete` `lock` `unlock`	Full object lifecycle
`source`	`read` `write` `check`	Source code with lock-safe writes
`test`	`run`	ABAP Unit test execution
`check`	`run`	ATC quality checks
`transport`	`list` `create` `release`	Transport request management
`ddic`	`table` `cds`	Table definitions and CDS view source
`package`	`list` `tree` `exists`	Package contents and tree traversal
`discover`	`services`	SAP system capability discovery

Every command works both as a CLI invocation (erpl-adt search query "ZCL_*") and as an MCP tool call. Same binary, same behavior.

Get Started

Download the latest release for your platform:

# Linux
curl -LO https://github.com/DataZooDE/erpl-adt/releases/latest/download/erpl-adt-linux-x86_64.tar.gz
tar xzf erpl-adt-linux-x86_64.tar.gz

# macOS (Apple Silicon)
curl -LO https://github.com/DataZooDE/erpl-adt/releases/latest/download/erpl-adt-macos-arm64.tar.gz
tar xzf erpl-adt-macos-arm64.tar.gz

Try the CLI — connect to your SAP system:

erpl-adt search query "Z*" --host sap.example.com --port 44300 --https --json

Configure as MCP server — add to your Claude Code or Gemini CLI config:

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

Then ask your AI agent anything about your SAP system. It will figure out which tools to call.

Star the repo: github.com/DataZooDE/erpl-adt

erpl-adt is part of the ERPL family of SAP integration tools by DataZoo. We build bridges between SAP and modern data & AI tooling.

Why This Exists​

What erpl-adt Does​

Demo: Claude Code Explores an ABAP System​

Same System, Different Agent: Gemini CLI​

What erpl-adt Handles for You​

The Full Command Set​

Get Started​