Local-first MCP context

Give agents exact APIs from your docs—not another web search

Budgeted, cited implementation packages from your own documentation, source trees, PDFs, and repositories.

Deliver more accurate agent output from your own corpus

ImplCache indexes documentation, project trees, PDFs, mirrored sites, and Git repos into SQLite on your machine—then returns the smallest sufficient package of implementation-ready context for the current task.

Exact Context

Replace noisy search with cited implementation packages

Primary tool get_implementation_context returns required APIs, symbols, sequences, examples, constraints, pitfalls, and citations—budgeted to your token limit.

Explore capabilities

Illustration of a budgeted implementation package with APIs, steps, pitfalls, and citation trails

Local-First

Keep the knowledge base on your machine

SQLite-backed, no cloud required. Optional Librarian UI manages sources, jobs, search, and health—embedded in the binary with no Node.js at runtime.

See how it works

Illustration of a local Librarian control surface managing sources, jobs, and search

Multi-Source Ingest

Index the corpora agents actually need

Ingest markdown docs, project trees, web mirrors, PDFs, and Git repos. Each corpus stays a portable URI tree so agents can choose roots instead of guessing across product families.

Start ingesting

Illustration of docs, code, PDFs, git repos, and sites flowing into a local knowledge store

Why it exists

Without ImplCache, agents scatter across files and the web. With it, they get exact symbols, project-local examples, and citations back to your corpus.

Side-by-side comparison: chaotic agent search on the left versus ordered cited packages on the right

Without ImplCache

Agent greps / opens many files
Agent skims broad docs
Agent searches the web
Agent may pick an obsolete API

            With ImplCache
            Agent gets exact symbols
One project-local example
Init order + constraints
Citations back to your corpus

          

What’s New, What’s NEXT

Staged retrieval depth—don’t dump the corpus. Start with a package, deepen only when needed.

01

get_implementation_context

Budgeted package: APIs, sequence, examples, constraints, pitfalls, citations.
02

find_symbol → search_knowledge

Drill into symbols across Go, C/C++/C#, Python, JS/TS, and Java extractors.
03

get_document

Pull the source document only when the package isn’t enough.

Local-first

SQLite on your machine; no cloud required.

Cited packages

Budgeted answers with source URIs.

Librarian UI

Dashboard, Sources, Jobs, Library, Search Lab, Health.

Agent & admin modes

Retrieval vs ingest / delete / recipes.

Built for Cursor and other MCP clients

Point your client at the binary over stdio, or run Librarian + HTTP on loopback. Ready-to-run packages ship for Windows, Linux amd64, and Jetson Orin NX—no Go install required.

Windows

dist/

Windows binaries, starter database, docs, and run-librarian helpers.

Open package on GitHub

Linux

dist-linux-amd64/

Linux amd64 package for desktops and servers—same Librarian UI and MCP agent mode.

Open package on GitHub

Jetson

dist-jetson-orin-nx/

NVIDIA Jetson Orin NX build for edge hosts; bind remotely for Cursor on a trusted network.

Open package on GitHub

MCP stdio and HTTP

3 packs Windows, Linux, Jetson

MIT Open source

Try it in 60 seconds

Choose your OS, grab the matching package, launch Librarian, then point Cursor at the binary.

1. Launch Librarian

cd dist
.\run-librarian.cmd
# open http://127.0.0.1:8080/

2. Point Cursor at the binary

{
  "mcpServers": {
    "implcache": {
      "command": "D:/path/to/dist/implcache-mcp.exe",
      "args": [
        "-db", "D:/path/to/dist/implcache.db",
        "-mode", "agent"
      ]
    }
  }
}

3. Ingest what you own

.\ingestcli.exe -db .\implcache.db `
  -mode markdown -root my-docs `
  -path "C:\path\to\docs"

.\ingestcli.exe -db .\implcache.db `
  -mode project -root my-app `
  -path "D:\work\my-app"

Get Windows package

1. Launch Librarian

cd dist-linux-amd64
chmod +x implcache-mcp ingestcli run-*.sh
./run-librarian.sh
# open http://127.0.0.1:8080/

2. Point Cursor at the binary

{
  "mcpServers": {
    "implcache": {
      "command": "/abs/path/dist-linux-amd64/implcache-mcp",
      "args": [
        "-db", "/abs/path/dist-linux-amd64/implcache.db",
        "-mode", "agent"
      ]
    }
  }
}

3. Ingest what you own

./ingestcli -db ./implcache.db \
  -mode markdown -root my-docs \
  -path /path/to/docs

./ingestcli -db ./implcache.db \
  -mode project -root my-app \
  -path /path/to/my-app

Get Linux package

1. Launch Librarian

cd dist-jetson-orin-nx
chmod +x implcache-mcp ingestcli run-*.sh
./run-librarian.sh
# LAN / remote Cursor:
./run-librarian-lan.sh
# open http://<jetson-host>:8080/

2. Point Cursor at the binary

{
  "mcpServers": {
    "implCacheRemote": {
      "url": "http://<jetson-host>:8080/mcp"
    }
  }
}

3. Ingest what you own

./ingestcli -db ./implcache.db \
  -mode markdown -root my-docs \
  -path /path/to/docs

./ingestcli -db ./implcache.db \
  -mode project -root my-app \
  -path /path/to/my-app

Get Jetson package Remote setup guide

Give agents exact APIs from your docs—not another web search

Deliver more accurate agent output from your own corpus

Replace noisy search with cited implementation packages

Keep the knowledge base on your machine

Index the corpora agents actually need

Why it exists

Without ImplCache

With ImplCache

What’s New, What’s NEXT

get_implementation_context

find_symbol → search_knowledge

get_document

Local-first

Cited packages

Librarian UI

Agent & admin modes

Built for Cursor and other MCP clients

dist/

dist-linux-amd64/

dist-jetson-orin-nx/

Try it in 60 seconds

1. Launch Librarian

2. Point Cursor at the binary

3. Ingest what you own

1. Launch Librarian

2. Point Cursor at the binary

3. Ingest what you own

1. Launch Librarian

2. Point Cursor at the binary

3. Ingest what you own

Learn ImplCache

Documentation

What is ImplCache

Implementation context

ImplCache vs RAG

Cursor MCP server

Guides