librocat

librocat

Provides AI agents with persistent memory over MCP by indexing OKF Markdown notes in a Git repository and exposing tools for searching, reading, writing, organizing, and reviewing knowledge.

Category
访问服务器

README

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. Install it into every agent you have — Claude Code, Cursor, Codex, VS Code, and others — with one command (the plugins CLI):

npx plugins add librocat/librocat

The plugin carries its own built server (server/main.js, zero dependencies, Node 20+), so the install downloads nothing from npm (the npm package librocat is a pointer that names the command above). To add just the MCP server to one client by hand, from a clone of this repo:

{ "mcpServers": { "librocat": { "command": "node", "args": ["<clone>/server/main.js"] } } }

Any runner runs the plugins CLI — npx, pnpm dlx, yarn dlx, or bunx. Skills alone: npx skills add librocat/librocat (skills CLI).

Set LIBROCAT_BUNDLE to your OKF directory (default ./okf). See docs/local.md.

The server runs on Node 20 or later and has zero runtime dependencies. It opens no network connection, makes no model call, and sends no telemetry: the agent that calls it is the AI. See SECURITY.md.

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)
plugin/         manifest sources for the plugin build (agent-plugins.org)
examples/       an example OKF bundle
docs/           Local-tier guides

Standards

  • 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.

推荐服务器

Baidu Map

Baidu Map

百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。

官方
精选
JavaScript
Playwright MCP Server

Playwright MCP Server

一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。

官方
精选
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。

官方
精选
本地
TypeScript
VeyraX

VeyraX

一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。

官方
精选
本地
graphlit-mcp-server

graphlit-mcp-server

模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。

官方
精选
TypeScript
Kagi MCP Server

Kagi MCP Server

一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

Exa MCP Server

模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。

官方
精选