@openerrand/mcp — OpenErrand MCP server

Lets an AI coding tool (Claude Code, Cursor, …) take a developer from a goal to a registered playbook + integration code — no recording, no hand-written JSON.

Tools

• playbook_guide — the authoring spec (actions, fence rules, inputs/templating)
• cold_start_playbook — draft a fenced, LLM-driven playbook from a goal (first-run recipe)
• lint_playbook — validate + danger-flag
• register_playbook — sign on-device (key kept in ~/.openerrand) + upload to the relay
• scaffold_integration — emit runnable client.run({ inputs, decide }) code

The customer's LLM orchestrates: read the guide → cold-start a fence from the goal → lint → register → scaffold. First runs are LLM-driven (the decider navigates the live page); deterministic caching/self-heal comes later.

Configure (Claude Code / Cursor)

Get your API key at https://openerrand.app/dashboard.html, then add:

{
  "mcpServers": {
    "openerrand": {
      "command": "node",
      "args": ["--import", "tsx", "/ABSOLUTE/PATH/openerrand/mcp/src/index.ts"],
      "env": { "OPENERRAND_API_KEY": "ak-..." }
    }
  }
}

Env: OPENERRAND_API_KEY (required), OPENERRAND_RELAY (default https://relay.openerrand.app).