Serve and connect

You have run your agent from the command line. Now you will start its HTTP server and talk to it over the network, the way a real client would. This continues from the my-agent project you built in the previous guides.

Start the server

The scaffolded manifest already defines an http_server block, so the binary can serve. Set your API key and start it:

export ANTHROPIC_API_KEY=your_key_here
./artifacts/build/my-agent serve

The agent starts, connects to its database, registers its tools, and begins listening on 127.0.0.1:8080. The process stays in the foreground, so leave it running and open a second terminal for the next step.

Send a request

The manifest enabled the AG-UI protocol, which exposes a /ag-ui/run endpoint. From your second terminal, send it a message:

curl -X POST http://localhost:8080/ag-ui/run \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [{ "role": "user", "content": "What is 21 plus 21?" }]
  }'

The response is a stream of Server-Sent Events. You will see the run start, the agent call the adder tool you added earlier, the streamed text of the answer, and finally the run finishing. That event stream is what a frontend consumes to render the conversation as it happens.

What you have done

You installed agentc, described an agent in a manifest, gave it a tool, compiled it to a single binary, and served it over HTTP. That is the full loop from an empty directory to a running agent.

Where to go next

• Author a manifest: build a manifest deliberately, block by block.
• Connect a CopilotKit frontend: put a UI in front of the AG-UI endpoint.
• Deploy a standalone binary: take the binary to production.
• HTTP reference: the native endpoints behind the server.