MidOS — MCP Server for Developer Knowledge

Curated, validated knowledge for AI coding agents. Not raw docs — battle-tested patterns.

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.

Quick Start

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

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

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

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

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)

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

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

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-mcp/
├── 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

Source-verified developer knowledge. Built by devs, for agents.
midos.dev · Discussions · Issues