ABAP MCP – AI-Powered SAP Development

An ABAP Model Context Protocol (MCP) server that brings AI-driven coding assistance to SAP ABAP development. Works with VS Code Copilot, Claude Desktop, Cursor, and any MCP-compatible client. Supports multiple LLM backends including GitHub Copilot, OpenAI, Anthropic Claude, Azure OpenAI, and local models via Ollama.

---

Architecture

┌──────────────────────────────────────────────────────────────────┐
│                       VS Code / IDE                              │
│                                                                  │
│  ┌─────────────────────┐    ┌──────────────────────────────────┐ │
│  │  GitHub Copilot Chat│    │  MCP Client (Claude Desktop,     │ │
│  │  @abap participant  │    │  Cursor, etc.)                   │ │
│  └──────────┬──────────┘    └──────────────┬───────────────────┘ │
│             │                              │                      │
│  ┌──────────▼──────────────────────────────▼───────────────────┐ │
│  │              VS Code Extension (abap-mcp-copilot)           │ │
│  │         • Copilot @abap chat agent                          │ │
│  │         • MCP client wrapper                                │ │
│  │         • Editor commands (analyze, format)                 │ │
│  └──────────────────────────┬────────────────────────────────--┘ │
└─────────────────────────────│────────────────────────────────────┘
                              │ stdio / SSE
┌─────────────────────────────▼────────────────────────────────────┐
│                     ABAP MCP Server (Node.js)                    │
│                                                                  │
│  Tools:                          Resources:                      │
│  • abap_analyze                  • abap://syntax-reference       │
│  • abap_generate                 • abap://best-practices         │
│  • abap_explain                                                  │
│  • abap_refactor                                                 │
│  • abap_search        ◄──────── SAP ADT REST API                │
│  • abap_unit_test                                                │
│  • abap_format                                                   │
│  • abap_docs                                                     │
│                                                                  │
│  ┌─────────────────────────────────────────────────────────────┐ │
│  │              Generic LLM Provider Layer                     │ │
│  │  ┌──────────┐ ┌──────────┐ ┌────────────┐ ┌─────────────┐ │ │
│  │  │  OpenAI  │ │Anthropic │ │Azure OpenAI│ │   Ollama    │ │ │
│  │  │  GPT-4o  │ │  Claude  │ │  GPT-4o   │ │  CodeLlama  │ │ │
│  │  └──────────┘ └──────────┘ └────────────┘ └─────────────┘ │ │
│  │                    ┌────────────────┐                       │ │
│  │                    │ GitHub Copilot │                       │ │
│  │                    └────────────────┘                       │ │
│  └─────────────────────────────────────────────────────────────┘ │
└──────────────────────────────────────────────────────────────────┘

---

Features

MCP Tools

Tool	Description
abap_analyze	Deep code analysis: syntax, performance, security, Clean ABAP, cyclomatic complexity. Optionally integrates SAP Code Inspector via ADT.
abap_generate	Generate production-ready ABAP from natural language – classes, CDS views, RAP behaviors, OData services, unit tests, and more.
abap_explain	Plain-English explanation of any ABAP code. Adjustable audience level (beginner → expert). Optional line-by-line annotations.
abap_refactor	Refactor towards Clean ABAP: extract methods, modernize syntax, improve naming, add error handling, convert to OOP.
abap_search	Search ABAP repository objects in a connected SAP system (requires ADT credentials).
abap_unit_test	Generate ABAP Unit Test classes using ABAP Unit + Test Double Framework. Optionally run existing tests via ADT.
abap_format	Format code: 2-space indentation, uppercase keywords, replace deprecated syntax (MOVE→=, SELECT *→explicit fields).
abap_docs	Comprehensive documentation for ABAP keywords, statements, classes, and SAP concepts with version-specific notes.

MCP Resources

Resource URI	Content
abap://syntax-reference	ABAP data types, modern 7.4+ syntax, constructor expressions, string templates
abap://best-practices	Clean ABAP naming, DB access patterns, exception design, unit testing

VS Code Extension (@abap Copilot Agent)

Use @abap in Copilot Chat with these slash commands:

@abap /analyze          ← analyze the current file
@abap /generate a CDS view for sales order header with customer join
@abap /explain          ← explain selected code or whole file
@abap /refactor         ← clean up current file
@abap /test             ← generate unit tests for current class
@abap /docs FOR ALL ENTRIES
@abap /format           ← format and fix deprecated syntax
@abap how do I implement a BAdi?   ← free-form ABAP question

LLM Providers

Provider	Config	Notes
GitHub Copilot	LLM_PROVIDER=copilot	Uses your existing Copilot subscription via VS Code
OpenAI	LLM_PROVIDER=openai + OPENAI_API_KEY	GPT-4o recommended
Anthropic	LLM_PROVIDER=anthropic + ANTHROPIC_API_KEY	Claude Sonnet 4.6 recommended
Azure OpenAI	LLM_PROVIDER=azure-openai + endpoint/key	Enterprise deployments
Ollama	LLM_PROVIDER=ollama	Local inference with CodeLlama, DeepSeek-Coder

---

Quick Start

1. Install dependencies

npm install

2. Configure environment

cp config/example.env .env
# Edit .env with your LLM credentials and optional SAP connection

3. Build and run

npm run build
npm start              # stdio mode (for Claude Desktop / Cursor)

4. Connect to Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "abap": {
      "command": "node",
      "args": ["/path/to/ABAP_MCP/dist/server/index.js"],
      "env": {
        "LLM_PROVIDER": "openai",
        "OPENAI_API_KEY": "sk-..."
      }
    }
  }
}

5. Install the VS Code Extension

cd vscode-extension
npm install
npm run compile
# Press F5 to launch Extension Development Host, or package with vsce

Then in VS Code settings, configure:

{
  "abapMcp.llmProvider": "copilot",
  "abapMcp.sapBaseUrl": "https://your-sap-system:8000",
  "abapMcp.sapUser": "DEVELOPER"
}

---

SAP System Connection (Optional)

The SAP connection uses the ABAP Development Tools (ADT) REST API, available on any SAP system with ABAP 7.5+ or S/4HANA.

Required SAP authorizations:

• S_ADT_RES – ADT resource access
• S_DEVELOP – ABAP Workbench authorization
• S_CODE_INS – Code Inspector (for runCodeInspector: true)

Set credentials:

SAP_BASE_URL=https://my-sap.example.com:8000
SAP_USER=DEVELOPER
SAP_PASSWORD=secret
SAP_CLIENT=100
SAP_LANGUAGE=EN

---

Development

npm run dev          # Run server in development mode (tsx, no build)
npm run build:watch  # Watch mode TypeScript compilation
npm run typecheck    # Type-check without building
npm test             # Run unit tests

Project Structure

ABAP_MCP/
├── src/
│   ├── server/
│   │   ├── index.ts              # MCP server entry point
│   │   ├── tools/                # MCP tool handlers
│   │   │   ├── analyze.ts        # abap_analyze
│   │   │   ├── generate.ts       # abap_generate
│   │   │   ├── explain.ts        # abap_explain
│   │   │   ├── refactor.ts       # abap_refactor
│   │   │   ├── search.ts         # abap_search
│   │   │   ├── unit-test.ts      # abap_unit_test
│   │   │   ├── format.ts         # abap_format
│   │   │   └── docs.ts           # abap_docs
│   │   └── resources/            # MCP resources
│   │       ├── abap-syntax.ts    # ABAP syntax reference
│   │       └── best-practices.ts # Clean ABAP guidelines
│   ├── sap/
│   │   ├── adt-client.ts         # SAP ADT REST API client
│   │   └── types.ts              # ABAP domain types
│   ├── llm/
│   │   ├── provider.ts           # Generic LLM interface + ABAP system prompt
│   │   ├── factory.ts            # Provider factory (reads env config)
│   │   ├── openai.ts             # OpenAI adapter
│   │   ├── anthropic.ts          # Anthropic adapter
│   │   ├── azure-openai.ts       # Azure OpenAI adapter
│   │   ├── copilot.ts            # GitHub Copilot adapter
│   │   └── ollama.ts             # Ollama (local) adapter
│   └── utils/
│       ├── config.ts             # Zod-validated env config
│       └── logger.ts             # Winston logger
├── vscode-extension/
│   ├── package.json              # VS Code extension manifest
│   └── src/
│       ├── extension.ts          # Extension activation, commands
│       ├── copilot-agent.ts      # @abap Copilot chat participant
│       └── mcp-client.ts         # MCP client wrapper
└── config/
    ├── example.env               # Configuration template
    └── mcp.json                  # MCP server config template

---

Adding a New LLM Provider

1. Create src/llm/my-provider.ts implementing LLMProvider:

export class MyProvider implements LLMProvider {
  readonly name = 'my-provider';
  readonly model: string;

  async complete(messages, options) { /* ... */ }
  async *stream(messages, options) { /* ... */ }
  async healthCheck() { /* ... */ }
}

2. Register it in src/llm/factory.ts under a new case.
3. Add the new provider to the LLM_PROVIDER enum in src/utils/config.ts.
4. Add a VS Code setting entry in vscode-extension/package.json.

---

License

MIT