Goals

1. Extend existing turborepo config with MCP support
2. Pack Turborepo caches and overall codebase as an LLM context, and update it incrementally according to the current state of the monorepo, for both one time and persistent actions
3. Filter out files out of LLM context according to turborepo config and .gitignore contents

Non-goals

1. Implement any form of LLM inference
2. Implement any form of Langchain RAG or LoRA / QLoRA tuning
3. Such solution should work with both with file patchsets out of git repo and concated files - there's no clear guarantee what would work better, especially for self-hosted llm's

Background

MCP
NX LLM Integration details

Proposal

Direct MCP turborepo client would provide direct LLM agents integration with Turborepo monorepo.
Simplifies incremental LLM-context updates and task output propagation, by reusing the existing caching capabilities.