Implement MCP support for Turborepo, to pack and manage LLM contexts, similarly to the existing NX AI offerings

[yuriy-yarosh]

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.

[orenmizr]

[yuriy-yarosh]

Oren, It does not really matter who's place this is, when you're not wrong and deserve to be heard.

From my superficial understanding, this whole situation with acqui-vacuuming things is just a sign of Potential React Cash Faucet Malfunction - it'll clog and Vercel needs a backup plan, when their offerings become more like convenience commodities, rather than convenience novelties to people.

Happens to any automation-focused business, especially open-source one.
We've already seen hashi-cluter-fudge of semi-monetized SaaS'Ses - deliberately planted engineering deficiency and design flaws to promote enter-pricy sales. And there's OpenNext.js for a reason...

[anthonyshew]

Hey, both of these comments are pretty weird. It's true that a company can both build and maintain existing software and purchase new software to build and maintain.

If you're running into a bug, please report it with a reproduction so we can fix it here: https://github.com/vercel/turborepo/issues/new?template=0-turborepo-bug-report.yml

[yuriy-yarosh]

@anthonyshew The issue in here is Agreeable Engineering Capacity and Agreeable Engineering Capability Thresholds, when certain engineering problems and design approaches have certain Stakeholders Agreed Threshold in both Quality, Quantity and Timeline of the actual Shipped Solutions... which actually undermines the Viability and Trust in OpenSource.

[anthonyshew]

Can you say what you're trying to accomplish with such a tool? What are the use cases?

[yuriy-yarosh]

There can be multiple Query Contexts and Multiple Agents assigned to a monorepo, parallelizing tasks in collaborative refactoring through shared planning.

[anthonyshew]

What problems can Turborepo uniquely and specifically solve for you? Can you provide a concrete example of where your LLMs are coming up short where Turborepo can provide context that will help direct it to its goals?

[ftzi]

https://nx.dev/blog/nx-mcp-vscode-copilot#why-use-nx-mcp-in-your-monorepo