██████╗ ██████╗  █████╗  ██████╗
 ██╔═══██╗██╔══██╗██╔══██╗██╔════╝
 ██║   ██║██████╔╝███████║██║  ███╗
 ██║   ██║██╔══██╗██╔══██║██║   ██║
 ╚██████╔╝██║  ██║██║  ██║╚██████╔╝
  ╚═════╝ ╚═╝  ╚═╝╚═╝  ╚═╝ ╚═════╝

ORAG is an organizational RAG and MCP platform that connects your internal data to AI pipelines — with enterprise access controls built in from day one.

Stop duct-taping retrieval pipelines together. ORAG gives your team a single place to index knowledge, expose it via the Model Context Protocol, and manage exactly who can access what — all with sub-50ms retrieval and zero infrastructure to babysit.

One config file is all it takes to expose your entire knowledge base as a fully typed MCP server — complete with streaming, auth, and observability out of the box.

Connect your org's documents, wikis, and databases into a unified retrieval layer. ORAG indexes, chunks, and embeds everything so your models always have the right context.

Spin up Model Context Protocol servers that give your agents structured, permissioned access to internal APIs, code repos, CRMs, and more — without bespoke glue code.

Role-based access across every knowledge base and MCP server. Audit logs, team workspaces, and SSO keep enterprise compliance teams happy.

Full request tracing across retrievals, tool calls, and completions. Latency breakdowns, token usage, and retrieval quality scores — all in one view.

01 → Connect sources      Link Notion, Confluence, S3, GitHub, or any custom data source.
02 → Configure pipelines  Choose chunking strategy, embedding model, and retrieval parameters.
03 → Expose via MCP       Your retrieval layer becomes a typed MCP server your agents can call.
04 → Ship to production   Deploy with one click. Monitor latency and quality from the dashboard.

One config block connects Claude, Cursor, or any MCP-compatible agent to your entire organization's knowledge base:

// ~/.claude/claude_code_config.json
{
  "mcpServers": {
    "orag": {
      "type": "http",
      "url": "https://orag.theanmolsharma.com/api/orgs/acme-corp/mcp",
      "headers": {
        "Authorization": "Bearer orag_live_xxxxxxxxxxxxxxxxxxxx"
      }
    }
  }
}

That's it. Your agents now have permissioned, streaming, observable access to everything your org knows.

All entities live in one DynamoDB table, keyed by pk + sk:

GSI1 — EMAIL#<email> → instant email lookup at sign-in
GSI2 — PUBLIC_ORGS → sparse index for public org discovery + search

Client                             Server
  │                                  │
  │── Google ID Token ─────────────▶ POST /api/auth/google
  │                                  ├─ Verify with Google
  │                                  ├─ Lookup via GSI1 (email)
  │                                  ├─ Create user if new
  │◀── JWT { userId, email, name } ──┤
  │                                  │
  │── Authorization: Bearer <JWT> ──▶ All subsequent API calls

// lib/middleware.ts
compose(withCors, withAuth)(handler)

// withAuth         → requires valid JWT, injects req.user or returns 401
// withOptionalAuth → injects req.user if present, never blocks
// withCors         → CORS headers + OPTIONS preflight

• Node.js 18+
• AWS account (DynamoDB)
• Google Cloud project (OAuth 2.0 credentials)

git clone https://github.com/Anmol202005/ORAG.git
cd ORAG
npm install

GOOGLE_CLIENT_ID=your_google_client_id
JWT_SECRET=your_jwt_secret

AWS_REGION=us-east-1
AWS_ACCESS_KEY_ID=your_key
AWS_SECRET_ACCESS_KEY=your_secret
DYNAMO_TABLE_NAME=your_table

# Frontend only (Vite dev server → localhost:5173)
npm run dev

# Full stack with API
npx vercel dev

Push to GitHub → import in the Vercel dashboard → add env vars → done.
The api/ directory is auto-detected as serverless functions. Zero config required.

ORAG/
├── api/                  # Vercel serverless functions
│   ├── auth/             #   google.ts · me.ts · signout.ts
│   └── orgs/             #   mine.ts · public.ts · [id]/join.ts · [slug].ts
├── lib/
│   └── middleware.ts     # withAuth · withOptionalAuth · withCors · compose
├── src/                  # React 18 frontend
│   ├── components/
│   └── pages/
├── vercel.json           # Deployment config
├── vite.config.ts
└── DynamoDoc.txt         # Full DB schema reference

1. Fork the repo
2. git checkout -b feat/your-feature
3. git commit -m 'feat: your change'
4. Open a Pull Request

Please run npm run lint before submitting.