Idf Mcp

Turns an Intent-Driven Framework domain into an MCP server where every tool description includes preconditions, invariants, irreversibility warnings, and role scopes. The runtime sits between Claude and your backend, rejecting bad actions before they hit storage with structured JSON responses the LLM can parse and adapt to. Tool descriptions show what can fail and why (like "Response: max 1 per taskId where status=selected"), and rejections come back as typed errors (preapproval_denied, no_preapproval) instead of generic 500s. You declare entities, intents, and invariants once in an IDF artifact, and the server derives all the guardrails. Reach for this when you're putting an agent into production and need domain-level constraints enforced before the call, not after.