You're an FDE at Antimetal. Customers want to give Antimetal access to their databases so our agent can connect and start analyzing their infrastructure. Each customer runs their own MCP server pointing at their DB and hands us the endpoint.
Acme Corp is the first — they've set up an MCP server and they're asking:
"We'd like to give Antimetal access. How should we send you credentials securely? And once you have them, how do we know the connection is actually working?"
The customer environment is already running in customer/. You don't need to touch it. Treat it as Acme's infrastructure.
To start it:
cd customer
cp .env.example .env
docker compose upThe MCP server is now running at: http://localhost:8000/sse
Your job is to build the Antimetal-side onboarding flow that gets a customer's MCP server connected and verified before the agent ever touches it.
-
Accept credentials securely Build a way for a customer to submit their MCP endpoint (and any credentials) to Antimetal. Persist them in a way you'd be comfortable with in production — Acme won't be the last customer.
-
Verify the connection Using the stored credentials, connect to the customer's MCP server, call a tool, and display a live result from their database in a simple web UI. This is the confirmation step — Acme should be able to look at it and know their data is coming through.
-
Hand off to the agent Once verified, the stored credentials should be ready for the Antimetal agent to pick up and use. You don't need to build the agent — just make sure what you've built gets it everything it needs.
Use whatever stack you're comfortable with.
- Credentials handled safely, with multiple customers in mind
- MCP tools callable from your backend using stored credentials
- Live query result visible in the UI
- Clean, readable code
1 hour. We'll walk through it together at the end.
- No need to README your own code — just write it
- Coding agents are fair game
- Ask questions if anything is unclear