zettcode-server

The ZettCode coding agent behind an HTTP API, in the same repository as the terminal program and its SDK.

One process serves one workspace. Every endpoint drives the released zettcode SDK — the same tools, plugins, MCP servers, skills, project instructions and session store as the terminal — so a conversation begun here can be reopened in zettcode, and the other way round. The package holds no agent of its own.

uv tool install zettcode-server
zettcode-server --workspace ~/projects/api

zettcode-server 0.1.0 (zettcode 0.1.8) · http://127.0.0.1:3290 · workspace /home/you/projects/api · permission deny · token off

It binds loopback on port 3290, denies shell commands by default, and asks for a bearer token only when one is configured (--token, $ZETTCODE_SERVER_TOKEN).

A machine with no model configured is still served: /health says ready: false and POST /models writes the [[models]] entry (the same one /config add writes) and builds the agent, so an editor can set ZettCode up without touching a file by hand.

curl http://127.0.0.1:3290/health
curl -N -X POST http://127.0.0.1:3290/prompts \
  -H 'content-type: application/json' \
  -d '{"text": "summarize the README", "approvals": {"mode": "deny"}}'

The stream carries the run's events and ends with its result. A run that needs an answer — a shell command, a question — is suspended and answered through POST /runs/approvals or POST /runs/questions, so a client stays in the conversation instead of guessing.

• The guide: install, endpoints, event stream
• TypeScript client: npm install zettcode-client
• Python SDK: the same agent from Python