What is ImplCache?

ImplCache is a local MCP (Model Context Protocol) server backed by SQLite. You ingest corpora you already trust—markdown docs, project trees, PDFs, mirrored sites, and Git repositories—then agents ask for implementation context for a specific coding task.

Instead of returning a pile of loosely related chunks, ImplCache aims to return the smallest sufficient package: required APIs, relevant symbols, a sensible sequence, examples, constraints, pitfalls, and citations back to your corpus.

The problem it addresses

When an agent works against an SDK or internal codebase, a few failure modes show up again and again:

• It greps widely and opens too many files.
• It skims broad documentation and misses init order or constraints.
• It searches the web and lands on obsolete APIs.
• It invents plausible but wrong call patterns.

ImplCache exists to keep the agent grounded in your material, with enough structure to write code that matches how the system actually works.

Who it is for

Operators and developers who already maintain product docs or source trees and want Cursor (or another MCP client) to reuse that material safely. It is especially useful when correctness depends on project-local examples, not generic internet advice.

What you run

• Agent mode — retrieval tools for coding sessions.
• Admin mode — ingest, delete, and corpus maintenance.
• Librarian UI — optional browser UI for sources, jobs, library, search lab, and health (embedded in the binary; no Node.js at runtime).

What it is not

ImplCache is not a general chat product, not a hosted RAG SaaS, and not a promise that symbol extraction is perfect. It is a local tool for budgeted, cited implementation packages. Schema and ranking can still change before 1.0—pin or vendor if you need hard stability.