Documentation

MCP server

Connect turbometrics directly to Claude Desktop and other AI tools — without copy and paste.

turbometrics can be connected directly to Claude and other AI tools — without copy and paste, without opening the dashboard. You ask questions in natural language, the AI tool queries turbometrics in the background and gives you a finished answer.

What you can do with it

Scans & performance

  • "What is the current score of turbopress.de?"
  • "Which findings does turbometrics.io have?"
  • "Has my score improved this week?"
  • "Compare turbopress.de with turbometrics.io"
  • "Scan https://new-domain.com and show me the results" (also works for new domains that are not monitored yet)
  • "Start a scan for example.com in the EU region"

Real user monitoring

  • "Show me the Core Web Vitals of turbopress.de"
  • "How has the LCP of my website developed over the last 30 days?"
  • "Which pages on turbopress.de have the worst LCP?"
  • "How did the INP value last week compare with this week?"

Alerts & account

  • "Show me all open alerts"
  • "List my latest scans filtered by status"
  • "Show me the details of alert #123"
  • "Mark all alerts as read"
  • "Which plan am I on, and how many API requests do I have left today?"

Requirements

  • A turbometrics account on the Starter plan or above
  • Claude, ChatGPT or another MCP-compatible tool

For Claude and ChatGPT you no longer need an API token — sign-in runs through turbometrics itself. A token is only needed for clients without OAuth support (see below).

Enter the server URL, sign in to turbometrics, confirm once — done.

  1. In Claude: Settings → Connectors → "Add custom connector"
  2. Enter the URL: https://turbometrics.io/mcp
  3. Click "Add", then "Connect"
  4. You land on a turbometrics page: sign in and click "Allow"

The connector applies across every device on your Claude account (claude.ai, desktop app, mobile apps).

You will find the connection afterwards under Profile → API in the "Connected apps" section, where you can revoke it at any time.

Note: Entering your own request headers is only available in Claude's connector UI through a beta that is not enabled for every account. The OAuth route above works regardless.

Setup in Claude Code

claude mcp add --transport http turbometrics https://turbometrics.io/mcp

The first call opens your browser for sign-in. No --header needed any more.

ChatGPT (developer mode)

Note: Developer mode is currently in beta and only available with ChatGPT Plus or Pro.

  1. In ChatGPT: enable Settings → Developer Mode
  2. Apps & Connectors → create a new connector
  3. Enter the server URL https://turbometrics.io/mcp
  4. Follow the turbometrics sign-in and confirm

Alternative: local setup via npx (mcp-remote)

For clients without OAuth support you can still integrate the MCP server locally via mcp-remote and an API token. Create the token under Profile → API.

Claude Desktop – Mac

  1. Install Node.js – one of the following methods:

    With Homebrew:

    brew install node
    

    Without Homebrew: nodejs.org/en/download → download and run the macOS installer (.pkg).

  2. Create an API token under Profile → API

  3. Open the configuration file:

~/Library/Application Support/Claude/claude_desktop_config.json
  1. Add the entry:
{
  "mcpServers": {
    "turbometrics": {
      "command": "/opt/homebrew/bin/npx",
      "args": [
        "mcp-remote",
        "https://turbometrics.io/mcp",
        "--header",
        "Authorization: Bearer YOUR_API_TOKEN"
      ]
    }
  }
}
  1. Restart Claude Desktop

Claude Desktop – Windows

  1. Install Node.js: nodejs.org/en/download → download and run the Windows installer (.msi).

  2. Create an API token under Profile → API

  3. Find the correct npx path — open CMD (Win + R, cmd, Enter) and type:

where npx

The path typically looks like this: C:\Program Files\nodejs\npx.cmd

  1. Open the configuration file. In the Explorer address bar, type:
%APPDATA%\Claude\claude_desktop_config.json
  1. Add the entry — use the path from step 3 for command, and replace YOUR_API_TOKEN:
{
  "mcpServers": {
    "turbometrics": {
      "command": "C:\\Program Files\\nodejs\\npx.cmd",
      "args": [
        "mcp-remote",
        "https://turbometrics.io/mcp",
        "--header",
        "Authorization: Bearer YOUR_API_TOKEN"
      ]
    }
  }
}

Note: Backslashes in JSON have to be written twice (\\).

  1. Restart Claude Desktop

ChatGPT Desktop

ChatGPT Desktop (Mac and Windows) supports MCP servers from version 1.2025.x. Setup is almost identical to Claude Desktop – only the path to the configuration file differs.

Requirement: Node.js must be installed (see the Claude Desktop section above).

Mac:

  1. Find the correct npx path:
which npx
  1. Open the configuration file (create it if necessary):
~/Library/Application Support/ChatGPT/mcp.json
  1. Add the entry – enter the path from step 1 and your API token:
{
  "mcpServers": {
    "turbometrics": {
      "command": "/opt/homebrew/bin/npx",
      "args": [
        "mcp-remote",
        "https://turbometrics.io/mcp",
        "--header",
        "Authorization: Bearer YOUR_API_TOKEN"
      ]
    }
  }
}
  1. Quit ChatGPT Desktop completely (Cmd+Q) and restart it.

Windows:

  1. Open the configuration file (create it if necessary). In the Explorer address bar, type:
%APPDATA%\ChatGPT\mcp.json
  1. Add the entry – use the npx path from where npx and enter your API token:
{
  "mcpServers": {
    "turbometrics": {
      "command": "C:\\Program Files\\nodejs\\npx.cmd",
      "args": [
        "mcp-remote",
        "https://turbometrics.io/mcp",
        "--header",
        "Authorization: Bearer YOUR_API_TOKEN"
      ]
    }
  }
}
  1. Quit ChatGPT Desktop completely (system tray → Quit) and restart it.

Note: If ChatGPT Desktop does not pick up the configuration, try mcp_servers.json as an alternative file name.

Other MCP clients

The MCP server can be used with all MCP-compatible tools, for example Cursor or Windsurf.

Setting Value
Server URL https://turbometrics.io/mcp
Transport Streamable HTTP
Auth OAuth 2.1 with PKCE — or Authorization: Bearer YOUR_API_TOKEN

The OAuth metadata lives at /.well-known/oauth-protected-resource and /.well-known/oauth-authorization-server. Clients identifying themselves via Dynamic Client Registration or a Client ID Metadata Document are both supported.

Available tools

Tool Description
list_domains All monitored URLs with their status
get_latest_scan The newest scan with score, findings, TTFB and a link to the report
get_scan_history Score history — for monitored and new domains
get_findings Detailed findings of a scan
list_alerts Open and resolved alerts
list_scans List scans — filterable by domain, status and page, with a link to the report
get_alert Details of a single alert
get_rum_summary Core Web Vitals p75 values (RUM)
get_rum_metric_history Daily history of a RUM metric (LCP, CLS, INP, FCP, TTFB)
get_rum_pages Slowest pages by metric
compare_domains Compare two domains directly, with links to both reports
trigger_scan Start a scan — for any URL, including new domains; optionally with region, force or auth
mark_alerts_read Mark alerts as read
get_account_info Account details: plan, API limits, RUM status

If the connection fails

The URL must be https://turbometrics.io/mcp — with .io. The endpoint also answers under turbometrics.de, but the sign-in does not live there. The client therefore cannot authorise itself and reports a connection error without saying why.

Older setups using a token in the request header no longer work. Remove the connector and set it up again — sign-in now goes through OAuth, and you will be taken through a consent dialog once.

For persistent errors, disconnect and reconnect. The access token is valid for one hour and is renewed in the background afterwards. Once a renewal has failed, the connector does not retry it on its own.

An error mentioning "503" is temporary. It means the sign-in could not be verified right now — not that it is invalid. Wait a moment and try again.

Plan

The MCP server is available from the Starter plan onward.

Further reading

REST APIPlans & limitsSource code on GitHubMCP registry entry