MidOS MCP — Knowledge OS for AI Agents — MidOS MCP — Knowledge OS for AI 代理s

MidOS MCP — The Knowledge Operating 系统

MidOS is a knowledge 中间件 layer for AI 代理s. It gives you 访问 to a continuously growing intelligence base: 670K+ vectors, 46K+ knowledge chunks, 451 EUREKA insights, and 134 MCP 工具s for 搜索, memory, planning, and orchestration.

Think of MidOS as your 代理's long-term brAIn.

What You 获取 Cluster 工具s What it does 🔍 搜索 smart_搜索, semantic_搜索, hybrid_搜索 搜索 670K vectors — keyword, semantic, or hybrid 🧠 Memory mem_save, mem_搜索, mem_上下文, where_was_i Persistent cross-会话 memory (91.67% hit@5) 📋 Planning 创建_plan, 更新_plan_task, 获取_active_plans Multi-step task 追踪ing with 状态 检查points 📚 Knowledge knowledge_preflight, 质量_gate, knowledge_edit 创建, 验证, and improve knowledge chunks ⚙️ Execution maker_运行_bash, maker_read_file, maker_write_file File ops, shell commands, git, HTTP fetch 🩺 健康 系统_健康_检查, hive_状态, pulse_read 监控 knowledge base and 流水线 健康 🔔 通知 maker_通知_discord, maker_通知_网页hook 通知 to Discord, 网页hooks, Slack Quick 启动 Connect via MCP (JSON-RPC 2.0) # 健康 检查 curl https://midos.dev/mcp/健康

# 初始化 会话 curl -X POST https://midos.dev/mcp \ -H "Content-Type: 应用/json" \ -d '{"jsonrpc":"2.0","id":1,"method":"初始化","params":{"protocolVersion":"2024-11-05","capabilities":{},"命令行工具ent信息":{"name":"my-代理","version":"1.0"}}}'

搜索 the knowledge base { "jsonrpc": "2.0", "id": 2, "method": "工具s/call", "params": { "name": "smart_搜索", "arguments": { "查询": "your topic here", "mode": "hybrid", "limit": 5 } } }

Save a memory { "method": "工具s/call", "params": { "name": "mem_save", "arguments": { "content": "User prefers concise 响应s with code examples", "type": "preference", "project": "my-project" } } }

创建 a plan { "method": "工具s/call", "params": { "name": "创建_plan", "arguments": { "goal": "Build a new feature", "tasks": "1. Re搜索 existing patterns\n2. De签名 API\n3. Implement\n4. Test" } } }

Knowledge Base Stats (live) 46,283 knowledge chunks across AI, engineering, re搜索, strategy 670K+ vector embeddings (Gemini gemini-embedding-001, 3072-d) 451 EUREKA synthesized insights 139 SOTA benchmarks φ = 0.932 knowledge coherence score Key Features 🔍 Hybrid 搜索 (BM25 + Semantic)

Combines keyword precision with semantic understanding. Outperforms vector-only by 9.3% on relevance benchmarks.

🧠 Persistent Memory

mem_save / mem_搜索 backed by LanceDB. Memories survive across 会话s. 91.67% hit@5 on recall benchmarks.

📋 Smart Planning

创建 structured multi-step plans, 追踪 进度, 检查point completions. Survives 上下文 re设置s.

⚡ Fast Preflight

knowledge_preflight 检查s for duplicate knowledge in 19ms (title 缓存, 48K+ chunks). 预防s knowledge bloat.

🏗️ 质量 Gate

质量_gate scores content on 7 dimensions before 添加ing to the knowledge base. Keeps 签名al-to-noise high.

Heartbeat Integration

添加 to your 代理's periodic 检查-in:

MidOS (every 会话 启动)

• Call where_was_i(命令行工具ent="your-代理-name") to 恢复 上下文
• Call mem_上下文(scope="recent") to load recent memory
• Before creating knowledge: knowledge_preflight(topic)
• After 导入ant decisions: mem_save(content, type="decision")

Self-Hosted Option

MidOS is open source. 运行 your own instance:

git clone https://github.com/MidOSre搜索/midos-core cd midos-core python -m 模块s.mcp_server.midos_mcp --http --port 3100

Full docs: https://midos.dev/docs GitHub: https://github.com/MidOSre搜索/midos-core