SAP RFC MCP Server

An enterprise-grade SAP RFC connector based on the Model Context Protocol (MCP). This project aims to provide AI agents (such as Claude Desktop, GPT-4, Cursor, etc.) with secure, efficient, and context-aware access to SAP systems. Through the standardized MCP protocol, agents can execute SAP RFCs and retrieve business data as if calling local functions.

🌟 Project Highlights

• Enterprise-Grade Security: Supports a Master Key management architecture with dynamic token generation, expiration management, and one-click revocation.

• Full-Chain Tracing: Identity-aware logging system where every log includes the client IP and agent identity, with automatic daily rotation at midnight.

• Intelligent Table Reading: Built-in data buffer protection mechanism that automatically handles DATA_BUFFER_EXCEEDED errors and supports iterative, secure reading of large-scale data.

• AI-Optimized Metadata: Provides detailed function interface definitions, parameter descriptions, and supports exporting knowledge bases in RAG-friendly (Retrieval-Augmented Generation) formats.

• High-Availability Sessions: Supports intelligent session reuse; agents do not need to manually manage Session IDs, as the system automatically restores active sessions based on tokens.

🚀 Core Capabilities (Tools)

This server exposes the following toolset to MCP clients, covering the entire process from system discovery to data operations:

1. Core RFC Interaction

• call_rfc_function: Dynamically calls any SAP RFC function, supporting parameter passing for complex nested structures.

• rfc_system_info: Retrieves metadata such as SAP system version, protocol, kernel, and database type.

• get_rfc_functions: Queries available remote call functions in the SAP system via fuzzy matching.

2. Secure Data Reading

• read_table: A table reading tool with automatic overflow protection that automatically selects fields based on the RFC 512-byte limit.

• read_table_complete: Iterative reading mode that supports requesting large numbers of fields exceeding buffer limits, with automatic batch processing.

• get_table_structure: Retrieves detailed definitions of SAP tables, including technical field attributes and multi-language descriptions.

• test_table_access: Quickly verifies the current user's read permissions and data connectivity for specific tables.

3. Metadata Management

• get_function_metadata: Retrieves the complete "manual" for a function, including parameter direction, type, length, and multi-language comments.

• search_rfc_functions: Performs full-text search on function names and descriptions to help agents discover appropriate interfaces.

• export_metadata_for_rag: Exports SAP interface knowledge as JSON for building offline knowledge bases for agents.

⚙️ Running Modes

The project supports three running modes to meet integration needs in different scenarios:

• HTTP Mode (Default): Based on streamable HTTP transport provided by FastAPI, supporting streaming (SSE) responses, suitable for remote agent access.

• STDIO Mode: Communicates via standard input/output, suitable for locally installed MCP clients like Claude Desktop.

• Dual Mode: Enables both HTTP and STDIO simultaneously, balancing local debugging and remote service needs.

📦 Quick Deployment

1. Environment Requirements

• Python: 3.9 or higher.

• SAP NW RFC SDK: Must install the official SAP SDK and configure the SAPNWRFC_HOME environment variable.

2. Installation

git clone <repo_url>
cd sap-rfc-mcp-server
python -m venv .venv
source .venv/bin/activate  # Windows 使用 .venv\Scripts\activate
pip install -e .

3. Configuration

Copy .env.example to .env and fill in your SAP connection parameters and security keys:

# SAP 连接
SAP_ASHOST=192.168.1.100
SAP_SYSNR=00
SAP_CLIENT=100
SAP_USER=YOUR_USER
SAP_PASSWD=YOUR_PASSWD

# 安全密钥
MCP_MASTER_KEY=admin-secret-key-2026

🔐 Authentication Guide

The server uses a two-layer authentication architecture:

1. Management Layer (Master): Uses MCP_MASTER_KEY to access /auth/* endpoints to perform token issuance.

2. Business Layer (Agent): Uses generated dynamic tokens to access /mcp/* endpoints to call SAP functions.

Example of generating a token (using Master Key):

POST /auth/token/generate
X-API-Key: admin-secret-key-2026
Body: {"agent_name": "Finance_Agent", "ttl_hours": 48}

📂 Project Structure

sap_rfc_mcp_server/
├── auth.py                # 动态 Token 管理与持久化
├── server.py              # MCP 工具定义与处理核心逻辑
├── sap_client.py          # 底层 pyrfc 连接管理与池化
├── rfc_table_reader.py    # 增强型表数据安全读取逻辑
├── streamable_http_server.py # HTTP/SSE 传输层实现
├── logger.py              # 身份感知的自动切分日志系统
└── metadata_manager.py    # SAP 元数据分析与 RAG 优化

📖 More Documentation

• QUICKSTART.md: 5-minute quick start guide.

• MODES_GUIDE.md: Detailed explanation of running modes and configuration.

• DEPLOY.md: Systemd and Docker deployment for production environments.

© 2026 SAP RFC MCP Server Team. Licensed under MIT.