OpenGraph MCP Server

What You Get

The OpenGraph MCP Server (opengraph-io-mcp) exposes every OpenGraph.io API endpoint to AI assistants through the MCP standard. This means your AI can:

• Unfurl URLs – Extract OpenGraph metadata, Twitter Cards, and page information from any website
• Capture Screenshots – Take high-quality screenshots at various viewport sizes
• Scrape HTML – Fetch raw page content with JavaScript rendering and bot detection bypass
• Extract Content – Pull specific HTML elements (headings, paragraphs) in structured format
• Query Pages – Ask questions about webpage content and get AI-powered answers
• Generate Images – Create illustrations, diagrams, icons, and social cards with AI

How It Works

The MCP server acts as a secure bridge between your AI assistant and the OpenGraph.io API:

┌─────────────────┐     MCP (JSON-RPC)     ┌──────────────────┐
│  AI Assistant   │ ──────────────────────▶ │  opengraph-io-mcp │
│  (Claude, etc.) │                         │    (local)        │
└─────────────────┘                         └────────┬──────────┘
                                                     │ REST API
                                                     ▼
                                           ┌──────────────────────┐
                                           │  api.opengraph.io    │
                                           └──────────────────────┘

Key benefits of this architecture:

• Credentials stay out of the AI model – whether you sign in or use an App ID, neither is ever sent to the AI model itself
• No code changes – The AI naturally calls the tools when relevant
• Works offline – If you run the server locally (stdio), it works without a network connection to mcp.opengraph.io

Available Tools

Data and image tools default to the current v3 API, which enables smarter rendering and proxy handling automatically.

Supported Clients

The OpenGraph MCP Server works with any client that supports the Model Context Protocol:

Security Model

The server can run two ways, each with its own security model:

Hosted (Streamable HTTP) — Recommended

• Sign in with your account – supported clients can authenticate through a secure browser sign-in instead of pasting a key. See Signing In.
• Or use an API key – pass your App ID as an x-app-id header if your client doesn't support the sign-in flow
• No persistent state – the server doesn't store any request data

Local (stdio)

• Runs on your machine – no network exposure; communication happens through standard input/output
• Environment variables – your App ID is stored in a config file, never transmitted to AI models
• No persistent state – the server doesn't store any request data

Transport Options

Streamable HTTP (Hosted) — Recommended

Point your client at the hosted server and you're done — no local install required:

{
  "mcpServers": {
    "opengraph": {
      "url": "https://mcp.opengraph.io/mcp"
    }
  }
}

If your client supports it, connecting with no headers triggers a browser sign-in automatically. Otherwise, add an x-app-id header with your App ID — see Signing In for both options.

Stdio (Local)

The MCP server runs as a subprocess of your AI client, communicating through stdin/stdout:

npx -y opengraph-io-mcp

Next Steps