learning-mcp-server

A minimal MCP server (three toy tools: add, echo, current_time) meant purely for learning how MCP servers deploy and run on SAP BTP's Cloud Foundry trial environment. No auth, no SAP backend calls — just get something running end to end.

Step 1 — Get a trial account

1. Go to https://www.sap.com/products/technology-platform/trial.html and sign up (SAP account required; free, no time-limited data beyond the 90-day trial window).

2. After signup you land in the SAP BTP cockpit with a subaccount and a Cloud Foundry space already created for you.

Keep in mind while you're learning: the trial account is for personal, non-production exploration only, resets after 90 days (or 30 days of inactivity), and has no SLA. That's all fine for this purpose.

Step 2 — Enable Cloud Foundry environment

1. In the cockpit, open your subaccount → Cloud Foundry Environment.

2. If it's not already enabled, click Enable Cloud Foundry, choose the free/trial plan, keep defaults, and create it.

3. Note the API Endpoint shown here (something like https://api.cf.us10-001.hana.ondemand.com) — you'll need it to log in.

4. A default space (usually called dev) is created automatically.

Step 3 — Install the Cloud Foundry CLI

Install cf from https://github.com/cloudfoundry/cli#installers-and-compressed-binaries (or via your package manager, e.g. brew install cloudfoundry/tap/cf-cli on macOS).

Verify:

cf --version

Step 4 — Log in

Trial accounts authenticate via SSO passcode rather than username/password:

cf login -a <API-ENDPOINT-FROM-STEP-2> --sso

This prints a URL — open it in your browser, copy the one-time passcode shown there, and paste it back into the terminal prompt. Then select your org and the dev space when prompted (or run cf target -o <your-org> -s dev afterward).

Step 5 — Deploy the server

From this project folder:

cf push

cf push reads manifest.yml, uploads the code, builds it with the Python buildpack, and starts it. When it finishes, it prints a route — something like:

routes: learning-mcp-server-<random>.cfapps.us10-001.hana.ondemand.com

That's your server's public HTTPS URL. Your MCP endpoint is at <that-url>/mcp.

Important for learning purposes only: this deployment has no authentication — anyone with the URL could call it. That's acceptable for a throwaway trial-account experiment, but never deploy an unauthenticated server like this with real data or real SAP system access behind it.

Step 6 — Test it

The easiest way to poke at it directly is the official MCP Inspector:

npx @modelcontextprotocol/inspector

This opens a local web UI. Choose Streamable HTTP as the transport, paste in https://<your-route>/mcp, connect, and you should see the three tools (add, echo, current_time) listed — click one to try it.

Step 7 — Connect it to Claude Desktop (optional)

Claude Desktop's config expects a local command, but remote Streamable HTTP servers can be bridged in via mcp-remote:

{
  "mcpServers": {
    "learning-server": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://<your-route>/mcp"]
    }
  }
}

Restart Claude Desktop and ask it to add two numbers or echo some text.

Useful commands while learning

cf logs learning-mcp-server --recent   # see recent logs (crash cause etc.)
cf logs learning-mcp-server            # tail live logs
cf apps                                 # list deployed apps and their routes/state
cf restart learning-mcp-server          # restart after config changes
cf delete learning-mcp-server           # tear it down, free up memory quota

Where to go from here

Once this deploys and responds correctly:

1. Add a tool that calls a real SAP API (e.g. a public S/4HANA sandbox OData service, no auth needed to start).

2. Then introduce a BTP Destination instead of a hardcoded URL.

3. Then add authentication (SAP Cloud Identity Services / XSUAA) in front of the server, following the security pattern from the earlier architecture discussion.

Each of those is a small, isolated next step — happy to scaffold any of them when you're ready.