Connect the MCP Server to Your LLM

Why Connect Now (and Not at the End)

Every step in this workflow exists twice: as a view in the visual graph interface, and as an MCP tool your LLM can call. Connecting the MCP server at the start means each insight page that follows gives you both routes — click in the app, or ask in the chat. It also lays the foundation for Step 4, where your graphs become the memory of your AI second brain.

MCP (Model Context Protocol) is the standard that lets LLMs use external tools. Unlike a raw API, the tools are self-describing: the model sees what each one does and combines them on its own. You ask “what are the content gaps in my notes?” — the model picks the right tool.

Option A — Claude (Recommended, 2 Minutes)

1. Get your API key from the InfraNodus API panel.
2. In Claude (web or desktop): Settings → Connectors → Add Custom Connector.
3. Enter the URL https://mcp.infranodus.com, name it “InfraNodus”, click Connect.
4. You’re redirected to InfraNodus to authorize with your API key (OAuth2). Done.

Option B — Claude Code

In the terminal, run (with your API key from the InfraNodus API panel):

claude mcp add infranodus -s user \
  -e INFRANODUS_API_KEY=YOUR_KEY \
  -- npx -y infranodus-mcp-server

-s user makes the server available in all your projects. Use -s project instead to write it to a shareable .mcp.json in the repo, or omit the flag to keep it local to the current project. Restart Claude Code and check the connection with /mcp.

Option C — ChatGPT

In ChatGPT: Apps page → Advanced Settings → enable Developer Mode → Create App, use the same URL, leave OAuth fields empty, connect, and paste your API key. Note: ChatGPT’s MCP support only works in Developer Mode and new conversations — Claude currently gives the smoother experience.

Option D — Local Clients (Cursor, Claude Desktop)

Add this to your client’s MCP config file — ~/.cursor/mcp.json for Cursor, claude_desktop_config.json for Claude Desktop (or ~/.claude.json if you prefer editing Claude Code’s config by hand):

{
  "mcpServers": {
    "infranodus": {
      "command": "npx",
      "args": ["-y", "infranodus-mcp-server"],
      "env": { "INFRANODUS_API_KEY": "your-api-key-here" }
    }
  }
}

Full per-platform instructions (n8n, Make, OpenClaw, self-hosting from the open-source repo) are in the deployment guides.

Test It

Paste any of these into your LLM chat:

• “Analyze this text and show me the main content gaps: …”
• “Create a knowledge graph from this text and save it to my InfraNodus account.”
• “Find my existing graphs where I talk about [topic].”
• “What are the content gaps in my graph called ‘[your graph name]’?”

If the model lists tools like generate_knowledge_graph or generate_topical_clusters and returns clusters and gaps, you’re connected.

Every insight page that follows gives you both routes: click in the app, or ask in the chat.

Privacy

The MCP server uses the API’s doNotSave mode by default for ad-hoc analyses, so texts aren’t stored in your account or logs unless you explicitly ask to save a graph. Most tools don’t call any external AI at all — the network analysis is InfraNodus’s own algorithms (details). According to our privacy policy and our terms of service, we don't use any of your data for AI training purposes.