ContactGraph
Sign In

Model Context Protocol

Connect your AI to your network.

ContactGraph is a remote MCP server. Point Claude, ChatGPT, Gemini, or any MCP client at it and your assistant can answer “who do I know at…” questions using your own enriched graph — no scraping, no API keys, just sign in with Google.

MCP server URL

https://api.contactgraph.ai/mcp

Streamable HTTP transport. A trailing slash is fine. This is an endpoint for agents — opening it in a browser will not show a page.

Authentication

OAuth 2.1 with dynamic client registration. No Client ID or Secret — your MCP client registers automatically and you sign in with Google.

First: create your free account and upload your contacts / LinkedIn connections. MCP connects an assistant to your graph, so there needs to be a graph to query. Get your graph.

Setup

Add ContactGraph to your client

Claude

Claude.ai and Claude Desktop, via custom connectors.

  1. In Claude, open Customize (top of the left sidebar) → Connectors → + → Add custom connector.
  2. Name it ContactGraph and paste the MCP URL above as the Remote MCP server URL.
  3. Leave Client ID / Secret empty (dynamic client registration), click Connect, and sign in with Google.
  4. Start a new chat, enable the ContactGraph connector, and ask e.g. “What VCs do I know?”

Note: If tools stop responding after a while, disconnect and reconnect the connector to refresh the token.

ChatGPT

Custom apps via Developer Mode (Plus, Pro, Business, Enterprise on the web).

  1. In ChatGPT Settings → Apps, open Advanced settings → Developer mode and turn it on.
  2. Click Create app, name it ContactGraph, paste the MCP URL, and choose OAuth (leave Client ID / Secret empty).
  3. Check “I trust this application”, click Create, and complete the Google sign-in.
  4. In a new chat, open the + menu → Developer mode → select ContactGraph, then ask your question.

Note: Write actions (sync, connect) ask for confirmation — review each tool call before approving.

Gemini CLI

Google’s open-source terminal agent, with automatic OAuth discovery.

  1. Run: gemini mcp add -s user --transport http contactgraph https://api.contactgraph.ai/mcp
  2. Start the Gemini CLI and run /mcp auth contactgraph, then sign in with Google in the browser.
  3. Verify with /mcp (tools appear as mcp_contactgraph_*), then ask your question.

Note: OAuth needs a local browser, so headless SSH sessions require port forwarding.

OpenClaw

Self-hosted agent gateway (WhatsApp, Telegram, and more).

  1. Install the skill: curl -s https://api.contactgraph.ai/skill.md -o ~/.openclaw/skills/contactgraph/SKILL.md
  2. Add contactgraph to your MCP config pointing at the MCP URL (the OAuth bridge plugin handles PKCE and refresh).
  3. Run /mcp auth contactgraph and sign in with Google; verify tools with /mcp tools.

Also works with Cursor, Claude Code, and any MCP-compatible client. For Gemini Enterprise and other advanced setups, see the full README.

How it works

What your agent does once connected

The server exposes a small set of tools. A typical run:

  1. connect_source — link Gmail (open the Google consent URL once)
  2. sync_source — import your contacts, then poll get_source_status until it reports partial or complete
  3. start_enrichment — resolve employers, roles, and social profiles (poll get_enrichment_status)
  4. query_network — ask questions like “Who do I know at Stripe?”

Then just ask, in plain language:

  • “Who do I know at Stripe?”
  • “What VCs have I met?”
  • “Find warm paths into AI startups that are hiring.”
  • “Who did I talk to about hiring?”

Terminal & custom agents

Prefer the raw API or a skill file?

Any agent that can make HTTP requests can use the same tool surface over REST, or load the published skill file for step-by-step instructions. Both live on the API host.

https://api.contactgraph.ai/skill.md
Read the skill fileView source on GitHub

Your network is yours. Point your agent at it.

Get your graph — freehello@contactgraph.ai

Nonprofit · Open source · Never shares your data · Free forever