librocat
OfficialProvides tools to manage persistent knowledge stored as OKF Markdown files in a Git repository, including searching, reading, writing, renaming, organizing, and reading the repository's Git history.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@librocatsearch my OKF notes for the design decision on authentication"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
librocat
Persistent AI memory, shipped as an Agent Plugin. Your AI agents forget everything when a session ends, so you repeat yourself every day, to every one of them. librocat catalogs your notes and files as Open Knowledge Format (OKF) Markdown and retrieves them for any agent over MCP. Tell them once. At its core it is a library and information science tool for organizing notes. Not a database: a library.
Knowledge lives as Open Knowledge Format (OKF) Markdown files in your own Git repo (GitHub, GitLab, or anywhere) — the repo is the source of truth. The librocat MCP server reads the files at launch into an in-memory search index (Orama) and gives your agent sixteen tools to search, read, link-walk, write, rename, weed, keep a tag thesaurus, write finding aids, and read the Git history of knowledge. Your agent is the cataloger: it writes titles, descriptions, and tags with your own AI. No database, no service, no account.
Want the same memory on every agent and machine, with a librarian? librocat Cloud is coming soon: Supabase for AI memory — the same tools, hosted. Libro, the library cat, will catalog what your agent leaves blank and file duplicates, retired concepts, and stray tags on a weekly review shelf for you to approve, with a revision on every write. Watch librocat.dev.
Use it
librocat is an Agent Plugin: the MCP server plus the Agent Skills, and this repo root is the plugin. Two install paths reach the same result — pick the one your agent supports.
Option A — the Agent Plugin (recommended)
One command installs the MCP server and the Agent Skills together, into every agent you have — Claude Code, Cursor, Codex, VS Code, and others — with the plugins CLI:
npx plugins add librocat/librocatThe plugin carries its own built server (server/main.js, zero
dependencies, Node 20+), so the install downloads nothing else from npm.
Any runner works the same way — npx, pnpm dlx, yarn dlx, or
bunx.
Option B — Skills + MCP, separately
Some agents don't yet support Agent Plugins but do support the two older, more widely adopted standards it's built from: Agent Skills (40+ clients) and MCP (nearly universal). Install both — the skills instruct the agent to call this server's tools, so skills alone do nothing and the server alone works without the cataloging guidance:
npx skills add librocat/librocat{ "mcpServers": { "librocat": { "command": "npx", "args": ["-y", "librocat"], "env": { "LIBROCAT_BUNDLE": "./okf" } } } }The npm package librocat is the server itself (npx -y librocat runs
it directly, zero dependencies). To run from a clone instead:
{ "mcpServers": { "librocat": { "command": "node", "args": ["<clone>/server/main.js"] } } }Set LIBROCAT_BUNDLE to your OKF directory (default ./okf). See
docs/local.md.
Same binary, same MCP config, for librocat Cloud
too: npx -y librocat login and paste a workspace token from the
dashboard's Connect page, and this process becomes the Cloud client on its
next start instead — no config to rewrite. logout, whoami, and
push (upload a local OKF bundle into the logged-in workspace once) round
out the CLI. Cloud's own code is not in this repository; login only
teaches this binary to talk to it.
The server runs on Node 20 or later and has zero runtime dependencies. With no token it opens no network connection, makes no model call, and sends no telemetry; with a token it opens exactly one, to your Cloud workspace. The agent that calls it is the AI. See SECURITY.md.
Related MCP server: Anamnesis
Develop
Node 24 and pnpm (npm install -g pnpm).
pnpm install
pnpm test # core suite + mcp end-to-end over stdio, against the built artifact
pnpm run build # bundle the server into packages/mcp/dist with esbuild
pnpm run plugin # assemble the Agent Plugin (plugin.json + mcp.json + skills)Releases are tagged and listed in CHANGELOG.md.
Layout
.github/ CI (install, lint, typecheck, build, test, plugin) and Dependabot
plugin.json the repo root IS the Agent Plugin (agent-plugins.org form;
mcp.json .plugin/plugin.json + .mcp.json mirror it for the plugins CLI)
skills/ Agent Skills (agentskills.io)
packages/core OKF read/write + the in-memory Orama index + the Service
packages/mcp the librocat MCP server (official MCP TypeScript SDK, stdio)
packages/npm the npm package `librocat`: `server/main.js` (the same
bundle as above) as its `bin`
plugin/ manifest sources for the plugin build (agent-plugins.org)
examples/ an example OKF bundle
docs/ Local-tier guidesStandards
Agent Plugins: https://agent-plugins.org
Agent Skills: https://agentskills.io
MCP (official TypeScript SDK): https://github.com/modelcontextprotocol/typescript-sdk
Apache-2.0. © 2026 Laughing Hermit, Inc. — librocat is a Laughing Hermit, Inc. product; https://librocat.dev is its only domain.
This server cannot be deployed
Maintenance
Related MCP Connectors
- TaprootOAuthcom.taproothq
Persistent memory layer for AI tools. Save and recall notes across Claude and other MCP clients.
Personal knowledge graph as an AI memory layer over MCP - read, save, and link your memories.
Persistent memory and knowledge management for AI agents with semantic search and 50+ tools.
MCP-native notes and memory for ChatGPT, Claude, and other AI tools.
Related MCP Servers
- AlicenseNot gradedqualityAmaintenanceEnables interaction with a local knowledge repository via MCP, providing tools for capturing, indexing, and searching Markdown notes with Git version control and optional semantic search.2 npmMIT
- AlicenseNot gradedqualityAmaintenanceProvides a local-first, file-based memory layer for Claude Code that syncs across machines via git over private networks, with MCP tools for search, write, and sync of human-readable markdown notes.Apache 2.0
- AlicenseAqualityBmaintenanceProvides cross-session persistent memory for coding agents via MCP tools to store, retrieve, and manage notes with hybrid keyword/semantic search and automatic deduplication.8151 npmMIT
- AlicenseNot gradedqualityBmaintenanceEnables AI assistants to interact with a Markdown-based knowledge base with backlinks, tags, and full-text search, supporting note creation, reading, updating, renaming, and deletion through MCP tools.10 npmMIT