MidOS Research Protocol

<!-- mcp-name: io.github.MidOSresearch/midos --> <p align="center"> <h1 align="center">MidOS — MCP Server for Developer Knowledge</h1> <p align="center">Curated, validated knowledge for AI coding agents. Not raw docs — battle-tested patterns.</p> </p>

<p align="center"> <a href="https://modelcontextprotocol.io"><img src="https://img.shields.io/badge/MCP-Compatible-blue?style=flat-square" alt="MCP Compatible"></a> <a href="https://claude.ai"><img src="https://img.shields.io/badge/Claude_Code-Ready-D79943?style=flat-square" alt="Claude Code"></a> <a href="https://cursor.com"><img src="https://img.shields.io/badge/Cursor-Ready-4B8BBE?style=flat-square" alt="Cursor"></a> <a href="https://github.com/cline/cline"><img src="https://img.shields.io/badge/Cline-Ready-green?style=flat-square" alt="Cline"></a> <a href="https://github.com/nicepkg/aide"><img src="https://img.shields.io/badge/Windsurf-Ready-purple?style=flat-square" alt="Windsurf"></a> <br> <a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-green?style=flat-square" alt="MIT License"></a> <a href="https://github.com/MidOSresearch/midos/stargazers"><img src="https://img.shields.io/github/stars/MidOSresearch/midos?style=social" alt="GitHub stars"></a> <a href="https://smithery.ai"><img src="https://img.shields.io/badge/Smithery-Listed-orange?style=flat-square" alt="Smithery"></a> <a href="https://www.python.org/"><img src="https://img.shields.io/badge/Python-3.10+-blue?style=flat-square&logo=python&logoColor=white" alt="Python 3.10+"></a> </p>

---

104 skill packs across 20+ tech stacks. 1,284 curated chunks. 104 validated discoveries. Every piece reviewed, cross-validated, and myth-busted.

Your agent asks: "How do I implement optimistic updates in React 19?"
MidOS returns: Battle-tested pattern with useOptimistic + Server Actions, validated Feb 2026.
Context7 returns: Raw React docs from reactjs.org.

Install

pip install midos

Quick Start

One line. Add to your MCP config and start querying:

<details> <summary><b>Claude Code</b> — <code>.mcp.json</code> or <code>~/.claude/settings.json</code></summary>

{
  "mcpServers": {
    "midos": {
      "url": "https://midos.dev/mcp"
    }
  }
}

</details>

<details> <summary><b>Cursor / Windsurf</b> — MCP Settings</summary>

Add a new server:

• Name: midos
• URL: https://midos.dev/mcp
• Transport: Streamable HTTP </details>

<details> <summary><b>Cline</b> — MCP Settings</summary>

{
  "mcpServers": {
    "midos": {
      "url": "https://midos.dev/mcp",
      "transportType": "streamable-http"
    }
  }
}

</details>

<details> <summary><b>Self-hosted</b> — Run locally</summary>

git clone https://github.com/MidOSresearch/midos.git
cd midos
pip install -e .
pip install -e hive_commons/
python -m modules.mcp_server.midos_mcp --http --port 8419

Then point your MCP client to http://localhost:8419/mcp. </details>

First Tool Call

After connecting, personalize your experience:

agent_handshake(model="claude-opus-4-6", client="claude-code", languages="python,typescript", frameworks="fastapi,react")

Then search for what you need:

search_knowledge("React 19 Server Components patterns")

Tools Reference

Community Tier (free, no API key)

Tool	Description	Example
search_knowledge	Search 1,284 curated chunks across all stacks	search_knowledge("FastAPI dependency injection")
hybrid_search	Combined keyword + semantic search with reranking	hybrid_search("PostgreSQL JSONB indexing")
list_skills	Browse 104 skill packs by technology	list_skills(stack="react")
get_skill	Get a specific skill pack (preview in free, full in Dev)	get_skill("nextjs")
get_protocol	Protocol and pattern documentation	get_protocol("domain-driven-design")
hive_status	System health and live statistics	hive_status()
project_status	Knowledge pipeline dashboard	project_status()
agent_handshake	Personalized onboarding for your model + stack	See example above

Dev Tier ($19/mo — full content + advanced search)

Tool	Description	Example
get_eureka	Validated breakthrough discoveries (104 items)	get_eureka("response-cache")
get_truth	Empirically verified truth patches (17 items)	get_truth("qlora-myths")
semantic_search	Vector search with Gemini embeddings (3072-d)	semantic_search("event sourcing CQRS")
research_youtube	Extract knowledge from video content	research_youtube("https://youtube.com/...")
chunk_code	Intelligent code chunking for ingestion	chunk_code(code="...", language="python")
memory_stats	Vector store analytics and health	memory_stats()
episodic_search	Search agent session history	episodic_search("last deployment issue")

Ops Tier (custom — security, infrastructure, advanced ops)

Contact for specialized knowledge packs. midos.dev/pricing

Skill Packs (104 and growing)

Production-tested patterns for:

Frontend: React 19, Next.js 16, Angular 21, Svelte 5, Tailwind CSS v4, Remix v2

Backend: FastAPI, Django 5, NestJS 11, Laravel 12, Spring Boot, Symfony 8

Languages: TypeScript, Go, Rust, Python

Data: PostgreSQL, Redis, MongoDB, Elasticsearch, LanceDB, Drizzle ORM, Prisma 7

Infrastructure: Kubernetes, Terraform, Docker, GitHub Actions

AI/ML: LoRA/QLoRA, MCP patterns, multi-agent orchestration, Vercel AI SDK

Testing: Playwright, Vitest

Architecture: DDD, GraphQL, event-driven, microservices, spec-driven dev

How MidOS is Different

	Raw Docs (Context7, etc.)	MidOS
Content	Documentation dumps	Curated, human-reviewed, cross-validated
Quality	No validation	5-layer pipeline: chunks → truth → EUREKA → SOTA
Search	Keyword matching	Semantic + hybrid search (Gemini embeddings, 3072-d)
Onboarding	Generic	Personalized per model + CLI + stack
Format	Raw text	Stack-specific skill packs with production patterns
Accuracy	Stale docs	Myth-busted with empirical evidence

Knowledge Pipeline

staging/ → chunks/ → skills/ → truth/ → EUREKA/ → SOTA/
 (entry)    (L1)      (L2)      (L3)     (L4)      (L5)

• Chunks (1,284): Curated, indexed knowledge across 20+ stacks
• Skills (104): Organized, actionable, versioned by stack
• Truth (17): Verified with empirical evidence
• EUREKA (104): Validated improvements with measured ROI
• SOTA (11): Best-in-class, currently unimprovable

Using an API Key

Pass your key via the Authorization header for Dev/Ops access:

{
  "mcpServers": {
    "midos": {
      "url": "https://midos.dev/mcp",
      "headers": {
        "Authorization": "Bearer midos_your_key_here"
      }
    }
  }
}

Get a key at midos.dev/pricing.

Architecture

midos/
├── modules/mcp_server/   FastMCP server (streamable-http)
├── knowledge/
│   ├── chunks/            Curated knowledge (L1) — 1,284 items
│   ├── skills/            Stack-specific skill packs (L2) — 104 items
│   ├── EUREKA/            Validated discoveries (L4) — 104 items
│   └── truth/             Empirical patches (L3) — 17 items
├── hive_commons/          Shared library (LanceDB vector store, config)
├── smithery.yaml          Smithery marketplace manifest
├── Dockerfile             Production container
└── pyproject.toml         Dependencies and build config

Tech Stack

• Server: FastMCP 2.x (streamable-http transport)
• Vectors: LanceDB + Gemini embeddings (22,900+ vectors, 3072-d)
• Auth: 3-tier API key middleware (community → dev → ops) with rate limiting
• Pipeline: 5-layer quality validation with myth-busting
• Deploy: Docker + Coolify (auto-deploy on push)

Contributing

MidOS is community-first. If you have production-tested patterns, battle scars, or discovered that a popular claim is false — we want it.

1. Search existing knowledge first: search_knowledge("your topic")
2. Open an issue describing the pattern or discovery
3. We'll review and add it to the pipeline

License

MIT

---

<p align="center"> Source-verified developer knowledge. Built by devs, for agents. <br> <a href="https://midos.dev">midos.dev</a> · <a href="https://github.com/MidOSresearch/midos/discussions">Discussions</a> · <a href="https://github.com/MidOSresearch/midos/issues">Issues</a> </p>