Self-hosted MCP server: your codebase as a knowledge graph, your secrets in an encrypted vault — both reachable by your AI assistant, and by nobody else.
Your knowledge. Your server. Your graph.
Every project you own becomes a searchable knowledge graph. Your API keys live in an encrypted vault. Claude — or any MCP client — gets both, and nobody else gets anything.
Website · Live benchmarks · Quick start · Architecture · Wiki
Every AI coding assistant has the same two blind spots.
It doesn't know your codebase. You paste files into the context window one at a time and hope you picked the right ones. Knowledge Hub maps each project into a graph of concepts, modules and documents — clusters, hub nodes, relations — and lets the assistant traverse it. Ask "how does a refund actually reach the ledger?" and it walks the graph instead of guessing.
It can't hold your secrets. So the API key gets pasted into the chat and lives in a transcript forever. Knowledge Hub keeps keys in an encrypted vault and gives the assistant a tool to fetch one when a task genuinely needs it — and every access lands in an audit log.
It runs on your machine. No third party ever sees your code or your keys.
Requirements: Linux, Python 3.12+, and graphifyy — a third-party graph engine (MIT, by Safi Shamsi). Knowledge Hub is the server, vault, scheduler, retrieval engine and UI around it.
The installer creates a virtualenv, generates your keys, installs a systemd user service plus the nightly timer, and opens a setup wizard in your browser: choose a password, pick the projects to map, pick an AI provider for the semantic pass — done. On its first query the hybrid engine downloads a local embedding model once (~470 MB); after that, retrieval is fully
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/BEKO2210--Knowledge-Hub.html)
Open-source, LLM-agnostic capability runtime for AI agents — 97 skills as typed, permission-enforced MCP tools. Final certification: 23/23 gates PASS.
HarjjotSinghh/locumDelegate Grok Bot coding work to your own local Claude Code / Codex CLIs via a custom MCP connector
azimov777/casefileThe task tracker your AI agents keep for each other: every task carries a case file, so the next agent picks up where the last one stopped. MCP server + live board. Self-hosted, one-line install.
korovin-aa97/talkthrough-mcpMCP server: turn narrated screen recordings into agent-ready data — local Whisper transcript, scene keyframes, OCR, wall-clock anchoring. Record your screen, talk — your AI agent files the bugs.
00200200/usagetrimKeep the signal. Cut the noise. Local CLI + MCP — fold verbose tool output for Claude Code, Codex, Cursor & Desktop.
siemd2/reel-relayShare an Instagram reel to Claude and have it watch, research, and build what it shows. Self-hosted MCP connector (yt-dlp + ffmpeg + Groq Whisper).
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.