Knowledge · MCP server
Knowledge-Engine
Point it at your markdown. You get full-text and semantic search, a local dashboard, and an MCP server so Claude or Cursor can query the whole corpus directly — running entirely on your own machine. It ships with a curated reference corpus already loaded.
- License
- MIT engine
- Runtime
- Python 3.12+
- Corpus
- 46 files 3 libraries
- Price
- $0 / $199 one-time
What it is
A knowledge base you run, not one you rent.
Most knowledge tools are someone else's cloud with your notes inside it. Knowledge-Engine is a process you run — your files, your machine, your search index, your call on whether anything ever leaves the building.
01
Full-text & semantic search
SQLite FTS5 gives instant keyword search the moment you index. Add an optional bge-m3 embedding pass for meaning-based search. No external vector database to stand up.
02
MCP-native, in the box
A JSON-RPC MCP server ships with the engine. Claude Desktop, Cursor, and Continue can query your corpus as a tool — not just open files. Most knowledge tools are not there yet.
03
Batteries included
It does not ship empty. A curated three-library reference corpus — 46 files across decision analysis, AI monetization, and system design — is loaded and searchable on the first run.
The bundled corpus
Curated, not scraped.
The three reference libraries are organized on a three-lens framework — every file is written to be read by an agent, not just a person.
FRAME → ANALYZE → DECIDE
Decision Analysis
Military and corporate frameworks — OODA, Cynefin, pre-mortem, red-teaming — for matching the right tool to the class of problem.
IDENTIFY → EVALUATE → EXECUTE
AI Monetization
Practical guidance on generating revenue with AI systems — business models, product development, the economics underneath.
FRAME → DESIGN → EVOLVE
System Design
Architecture work products — blueprints, workflows, data contracts, roadmaps — for deciding what to build and how the parts fit.
Who it's for
If this sounds like you, it will fit.
- A developer or small team with a markdown corpus — notes, docs, research — that has grown past what
grepcan handle. - Anyone running Claude Desktop, Cursor, or Continue who wants those agents to search a real knowledge base, not just whatever file is open.
- People who will not put their notes in someone else's cloud and want search that runs on their own hardware.
Plainly
What it does, and what it doesn't.
What it does
- Indexes your markdown into a searchable SQLite database
- FTS5 keyword search, instant on index
- Optional bge-m3 semantic search over the same content
- Serves an MCP tool endpoint for Claude, Cursor, Continue
- Runs a local FastAPI dashboard for browsing and search
- Ships the curated three-library corpus, ready on first run
What it doesn't
- It is not a hosted service — there is no SaaS to log into
- Semantic search needs a local Ollama install; the model download is on you
- It is not multi-user auth — v1.0 assumes a local-trust machine
- It is not a Pinecone or Qdrant replacement — it is FTS5 plus cosine over SQLite
- It will not write your notes for you — it indexes what you already have
Get it
Two ways in.
Start free and self-host the engine. The Standard build adds the curated corpus and the rest of the batteries.
Free core
Core
The engine itself, cloned from a public template.
- FTS5 indexer + local dashboard
- JSON-RPC MCP server
- Optional bge-m3 embedding search
- Sample content to start from
- Point it at your own corpus
No account. Clone it and it is yours.
Batteries included
Standard
The engine, plus the curated corpus and the extras.
- Everything in Core
- The curated three-library corpus — 46 files
- Additional sample skills
- Dashboard themes
- Buyer update channel
One-time price, bought through Polar.sh. 30-day no-questions refund.
Run your knowledge base on your own machine.
Clone the free core, point it at your markdown, and wire the MCP server into your agent. Decide on the Standard build once you have seen it work.