🚀 SAP BTP MCP Server

A professional Model Context Protocol (MCP) server providing a robust natural language interface for SAP Business Technology Platform (BTP).

Manage your cloud infrastructure, security, and entitlements conversationally through AI agents (Claude, Gemini, etc.) using your local btp CLI session.

🌟 Key Features

• Conversational Infrastructure: Provision subaccounts and assign roles using simple language.

• Fail-Safe Design: Built-in timeouts, non-interactive execution, and resilient output parsing.

• Secure by Design:

  • Runs entirely locally; no external credentials or tokens are stored.

  • Leverages your existing BTP CLI authentication (SSO/JWT).

• Proactive Error Handling: Provides clear instructions when authentication expires or inputs are invalid.

• Automatic Discovery: Intelligently locates the btp binary across Windows, macOS, and Linux.

Related MCP server: SAP OData to MCP Server

🏗️ Fail-Proof Architecture

This server is designed for mission-critical AI automation:

• Intelligent Timeouts: Prevents "hanging" on long SAP BTP API operations with 60s/120s thresholds.

• Deep JSON Recovery: Automatically extracts valid data even if the CLI returns mixed output or warnings.

• Input Validation: Pre-validates technical IDs (GUIDs) and Emails to prevent unnecessary CLI failures.

• Non-Interactivity: Guaranteed non-blocking execution using CI-mode environments and null-input piping.

📋 Prerequisites

1. SAP BTP CLI: Installed and in your system PATH.

   • Download here

2. Active Session: You must be logged in. Verify with: btp list accounts/global-account

3. Python 3.10+: Ensure a modern Python environment.

🛠️ Installation

Using uv (Recommended)

uv tool install --from . btp-mcp-server

Using pip

pip install .

⚙️ Configuration

Add the following to your MCP client config (e.g., claude_desktop_config.json):

Using uv (Fastest)

{
  "mcpServers": {
    "btp": {
      "command": "uv",
      "args": [
        "--directory",
        "/absolute/path/to/BTP MCP Server",
        "run",
        "btp-mcp-server"
      ]
    }
  }
}

Using Python

{
  "mcpServers": {
    "btp": {
      "command": "python3",
      "args": ["-m", "btp_mcp_server"],
      "env": { "PYTHONPATH": "/absolute/path/to/BTP MCP Server/src" }
    }
  }
}

🧠 Available Tools

💬 Usage Examples

• "Check if my BTP session is still active using ping."

• "Show me all subaccounts in region eu10."

• "Assign the 'Subaccount Admin' role to colleague@example.com."

• "Create a development subaccount named 'Internal-Alpha' in us10."

🆘 Troubleshooting

Authentication Error (❌ AUTHENTICATION ERROR) If you see this, your BTP CLI session has expired.

1. Clear the error in the AI chat.

2. Run btp login in your terminal.

3. Refresh the AI session.

CLI Not Found (⚠️ BTP CLI not found) Ensure btp is in your PATH. On macOS/Linux, try ln -s /path/to/btp /usr/local/bin/btp.

🏗️ System Design

The server follows a layered architecture to ensure separation of concerns and maximum stability:

• Tool Layer (server.py): Uses FastMCP to register Python functions as MCP tools. Implements strict Pydantic validation and maps technical exceptions to human-readable markdown tips.

• Service Layer (btp_cli.py): Orchestrates command execution. Manages the state of the CLI path, handles binary auto-discovery, and implements the "Deep JSON Recovery" algorithm.

• Execution Layer (subprocess): Interacts directly with the OS. Uses hardened environments (CI=true) and standard error redirection to maintain security and non-interactivity.

• Utility Layer (utils.py): Centralizes cross-cutting concerns like logging to stderr and custom domain-specific exceptions.

🔄 Control Flow

1. Request Intake: The AI client (e.g., Antigravity) sends a JSON-RPC request to a registered tool (e.g., btp_list_subaccounts).

2. Pre-Validation: The Tool Layer hydrates parameters and performs regex-based technical validation (IDs/GUIDs).

3. Execution: The Service Layer constructs a safe CLI command, injecting --format json as a global prefix.

4. Retry & Recovery:

   • If the command times out, it retries with exponential backoff.

   • If the output contains warnings + JSON, the "Deep Recovery" engine extracts the valid payload.

5. Error Mapping: If the CLI returns "Session Expired", the decorator catches it and provides the user with an exact btp login command for their specific binary path.

6. Structured Response: The final JSON payload is beautified and returned to the AI as a markdown-formatted string.

📄 License

MIT License.

👤 Author

Kallol Chakraborty - @kallolchakraborty