/ai--mcp-server

Designs a system of several AI agents with an orchestration pattern, contracts between agents and failure handling, or advises a single agent instead.

Reviews and rewrites your system prompts, user prompts and tool definitions, then gives an A/B comparison of the original and improved versions.

Builds a test harness that measures a RAG pipeline: a golden question set, retrieval metrics, groundedness checks and a regression gate.

Adds a cache in front of LLM calls so repeated or near-duplicate requests get a stored answer, and measures whether it helps without serving wrong ones.

Makes a model return dependable JSON or typed objects: schema design, native structured modes, validation and bounded repair loops.

Audits or implements the tools a model can call: clear schemas, parallel calls, error handling that lets the model recover, and an accuracy eval.