Claude Agent
Run the Claude Code agent in a session using an Anthropic API key.
Run the Claude Code agent inside a VM session and drive it with prompts. Reach for this when you want a coding agent that reads, writes, and runs commands in an isolated environment instead of calling the model API directly.
How it works
The server registers a VM with the Claude Code software and starts the registry. The client creates a session against the claude agent, passing ANTHROPIC_API_KEY through the session env, then calls sendPrompt to get the agent’s response. From there you can layer on extras: drop a SKILL.md into ~/.claude/skills/ before creating the session and the agent discovers it automatically, or pass mcpServers (local child processes or remote URLs) to expose more tools. Pre-install local MCP servers with exec so first-run npx output does not corrupt the stdio handshake.
Run it
npm install
ANTHROPIC_API_KEY=sk-ant-... npm run start
The agent answers the prompt and prints its response to the console.