SAP BTP Services Discovery — MCP Server

A production-ready MCP (Model Context Protocol) server that exposes SAP BTP platform APIs as tools for any MCP-compatible AI agent — Claude Code, Cursor, Joule Studio, LangGraph agents, or any other MCP client.

Connect any AI agent to your SAP BTP landscape via the Model Context Protocol.

What is this?

The BTP MCP Server is an open-source Model Context Protocol server that connects AI agents to the SAP Business Technology Platform (BTP) APIs.

Instead of switching to BTP Cockpit to answer common developer questions, you can ask your AI agent directly:

• "What BTP services are available in my account?"
• "Is HANA Cloud already running in my subaccount?"
• "I need async messaging in my CAP app — what should I use?"
• "What destinations are configured and what authentication do they use?"

Works with Claude Code, Cursor, Joule Studio, LangGraph agents, and any other MCP-compatible AI client.

Tools

Installation

pip install btp-mcp-server

Prerequisites

You need a Service Manager service key from your BTP subaccount.

Steps to get one:

1. Go to BTP Cockpit → your subaccount → Services → Service Marketplace
2. Search for Service Manager → Create instance with plan subaccount-admin
3. Create a Service Key on the instance
4. The key JSON contains your credentials — see Configuration below

Configuration

All configuration is via environment variables. Copy .env.example to .env and fill in the values from your Service Manager service key:

# From your Service Manager service key JSON
BTP_CLIENT_ID=your-client-id          # from "clientid"
BTP_CLIENT_SECRET=your-client-secret  # from "clientsecret"
BTP_TOKEN_URL=https://your-subdomain.authentication.us10.hana.ondemand.com/oauth/token
BTP_SM_URL=https://service-manager.cfapps.us10.hana.ondemand.com  # from "sm_url"

# From BTP Cockpit → your subaccount → Overview → "Subaccount ID"
BTP_SUBACCOUNT_ID=your-subaccount-guid

# Destination Service URL (region-specific — adjust region if needed)
BTP_DESTINATION_URL=https://destination.cfapps.us10.hana.ondemand.com

# Optional: cache TTL in seconds (default: 300)
# CACHE_TTL_SECONDS=300

Note: Never commit your .env file to GitHub. Use .env.example (with no real values) as a reference template.

Usage

With Claude Desktop

Add to your claude_desktop_config.json:

Mac: ~/.claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "sap-btp": {
      "command": "btp-mcp-server",
      "env": {
        "BTP_CLIENT_ID": "your-client-id",
        "BTP_CLIENT_SECRET": "your-client-secret",
        "BTP_TOKEN_URL": "https://your-subdomain.authentication.us10.hana.ondemand.com/oauth/token",
        "BTP_SM_URL": "https://service-manager.cfapps.us10.hana.ondemand.com",
        "BTP_SUBACCOUNT_ID": "your-subaccount-guid",
        "BTP_DESTINATION_URL": "https://destination.cfapps.us10.hana.ondemand.com"
      }
    }
  }
}

Restart Claude Desktop. Then ask:

• "What BTP services do I have available?"
• "Are there any failed service instances?"
• "I need to connect to an on-premise SAP system — what BTP service handles that?"

With Claude Code / Cursor

{
  "mcpServers": {
    "sap-btp": {
      "command": "btp-mcp-server",
      "env": {
        "BTP_CLIENT_ID": "...",
        "BTP_CLIENT_SECRET": "...",
        "BTP_TOKEN_URL": "...",
        "BTP_SM_URL": "...",
        "BTP_SUBACCOUNT_ID": "..."
      }
    }
  }
}

With a LangGraph agent

from langchain_mcp_adapters.client import MultiServerMCPClient
from langgraph.prebuilt import create_react_agent
from langchain_anthropic import ChatAnthropic

async with MultiServerMCPClient({
    "btp": {
        "command": "btp-mcp-server",
        "transport": "stdio",
        "env": {
            "BTP_CLIENT_ID": "...",
            "BTP_CLIENT_SECRET": "...",
            "BTP_TOKEN_URL": "...",
            "BTP_SM_URL": "...",
            "BTP_SUBACCOUNT_ID": "...",
        }
    }
}) as client:
    tools = client.get_tools()
    agent = create_react_agent(
        ChatAnthropic(model="claude-sonnet-4-6"),
        tools
    )
    result = await agent.ainvoke({
        "messages": [{"role": "user", "content":
            "What BTP services do I have? Is there anything I should use for messaging?"
        }]
    })

With Joule Studio (BTP)

Deploy in HTTP mode and register as a BTP Destination:

MCP_TRANSPORT=http btp-mcp-server
# Server starts at http://0.0.0.0:8080/mcp

Then in BTP Cockpit → Destinations → create a destination pointing to your server URL. In Joule Studio Agent Builder → Tools → Add MCP Server → select the destination.

Running locally from source

git clone https://github.com/ABRANJAN07/btp-mcp-server.git
cd btp-mcp-server

pip install -r requirements.txt
cp .env.example .env
# fill in .env with your BTP credentials

# Test BTP connectivity
python test_connection.py

# Start the MCP server
python server.py

Running tests

Tests use mocked BTP responses — no real credentials needed.

pip install -r requirements.txt
pytest tests/ -v

Expected output: 17 tests passed

Project structure

btp-mcp-server/
├── server.py              ← MCP server entry point (5 tools)
├── test_connection.py     ← Verify BTP connectivity
├── requirements.txt
├── .env.example           ← Configuration template
├── btp_mcp/
│   ├── config.py          ← Reads .env settings
│   ├── auth.py            ← OAuth2 token management
│   ├── btp_client.py      ← BTP API calls + caching
│   ├── cache.py           ← TTL cache
│   └── models.py          ← Pydantic response models
└── tests/
    └── test_tools.py      ← 17 tests with mocked responses

How it works

AI Agent (Claude / Cursor / LangGraph)
    │
    │  MCP Protocol (stdio)
    ▼
BTP MCP Server (this package)
    │  OAuth2 client_credentials
    │  + TTL cache (5 min)
    ▼
SAP BTP APIs
  ├── Service Manager API  → service catalog, instances, plans
  └── Destination API      → configured connections

Caching: BTP API responses are cached for 5 minutes by default. The service catalog and instance list change rarely — caching keeps responses fast without sacrificing data freshness.

Pagination: All list endpoints fetch every page from BTP, not just the first 50 results.

Roadmap

This is Phase 2 of a 6-phase project building toward a full AI-powered BTP operations platform.

Coming soon (Phase 3)

• LangGraph ReAct agent with multi-turn conversation memory
• FastAPI SSE streaming endpoint
• Code generation for recommended services (CAP binding config, CLI commands, YAML manifests)

Entitlements API (temporarily disabled)

The get_btp_entitlements tool requires a separate Cloud Management Service (CIS Central plan) service key. The code is fully written and commented out — see ENTITLEMENTS_SETUP.md to enable it when ready.

Contributing

Contributions are welcome! Please open an issue first to discuss what you'd like to change.

git clone https://github.com/ABRANJAN07/btp-mcp-server.git
cd btp-mcp-server
pip install -r requirements.txt
pytest tests/ -v   # make sure tests pass before submitting a PR

License

MIT — see LICENSE for details.

Author

Built by Abhijeet Ranjan as part of a series on AI + SAP BTP integration.

Follow the journey on LinkedIn for Phase 3 updates.