Munin AI Memory

Munin AI Memory

Munin is a high-performance, pragmatic memory layer for AI agents (Cursor, Claude Code, OpenClaw, Gemini CLI,...). Unlike other solutions, Munin focuses on developer productivity with: * Multi-Project Support: Isolate memories into separate "brains" (Context Cores). * GraphRAG: Automatically builds a knowledge graph from your context. * Sub-200ms Search: Blazing fast Hybrid & Semantic

Category
访问服务器

README

🧠 Munin Ecosystem for AI Agents

Status: Active Powered by: GraphRAG Protocol: MCP License: MIT

Give your AI Agents a robust, Long-Term Memory.

Have you ever been frustrated when your AI agent forgets the architectural decisions you made yesterday? Or when it repeats the exact same bug it fixed in the previous session?

Munin is a Full-Stack Long-Term Memory manager powered by GraphRAG. This monorepo contains the official Model Context Protocol (MCP) adapters and SDKs to connect Munin Context Cores to your favorite AI tools—allowing them to build, query, and maintain a persistent knowledge graph of your entire project across endless sessions.


✨ Feature Highlights

Munin isn't just a database; it's a cognitive layer for your AI agents:

  • 🛡️ AI Memory Guard: Detects semantic contradictions in your agent's memory to ensure consistency.
  • 🕸️ GraphRAG Visualizer: Auto-extracts entities and relationships into interactive 2D neural knowledge graphs and Mermaid-compatible diagrams.
  • ⚡ Lower Token Costs: Semantic hybrid search (Vector + Keyword) ensures agents pull only the most relevant snippets, keeping prompts lean and fast.
  • 🔐 E2EE With GraphRAG: Industry-leading security. Encrypt your memory end-to-end while maintaining the ability to perform high-performance semantic search (Elite Tier).
  • 🕒 Temporal Search: Search by time context—ask "what did we decide last Tuesday?" and get exact answers.
  • 📌 Dynamic Pinning: Force-inject critical project context (like coding standards or core architecture) so AI never loses the "big picture".
  • 🤝 Cross-Project Sharing: Share selected memories across different projects to reuse logic and context without manual copy-pasting.
  • ⌛ Memory TTL: Set expiration windows for temporary context to keep your memory cores clean and noise-free.

🔌 Supported Adapters

This ecosystem provides first-class, plug-and-play MCP adapters for the most popular AI development tools. Choose your platform to get started:


📦 Monorepo Structure

This repository is organized as a pnpm workspace containing the core SDKs, the protocol specification, and all individual adapters:

  • Protocol Spec: packages/spec
  • TypeScript SDK: packages/ts-sdk
  • Python SDK: packages/python-sdk
  • First-Class Adapters: adapters/*
  • Generic MCP Template: adapters/generic-mcp-template
  • Contract Test Harness: tests/contract
  • Release Tag Mapping: docs/release-tags.md

🛠️ Developer Guide

If you are contributing to the Munin Ecosystem, use the following commands to manage the monorepo.

Quick Commands

pnpm install
pnpm lint
pnpm build
pnpm test
pnpm test:contract

Contract Test

Start the mock server (default 4010):

pnpm test:contract:mock

If the port is occupied, run on another port:

MUNIN_CONTRACT_PORT=4011 pnpm test:contract:mock
MUNIN_CONTRACT_PORT=4011 pnpm test:contract

You can also override the full base URL directly:

MUNIN_CONTRACT_BASE_URL=http://127.0.0.1:4011 pnpm test:contract

By default, the contract runner uses:

  • tests/contract/adapter-manifests/munin-sdk-local.json

Override with a custom manifest:

pnpm test:contract -- tests/contract/adapter-manifests/<manifest>.json

Built with ❤️ by Kalera for the AI Engineering community.

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