How to Connect Claude to GoHighLevel

You connect Claude to GoHighLevel by adding an MCP server URL to Claude's configuration and authorizing it with a GoHighLevel credential. There are two servers you can point at, and the setup steps are nearly identical for both — what differs is the URL and the credential.

They are not exclusive — MCP clients hold several servers at once and Claude picks per request. If you are only deciding where to start, start with HighLevel's; the comparison page sets out where each is stronger.

A Private Integration Token (PIT) is GoHighLevel's long-lived API credential for a single sub-account. Per HighLevel's MCP documentation, create one from Settings → Private Integrations inside the location you want to use, click Create New Integration, select the scopes, and copy the generated token. It begins with pit-.

Scopes are the security boundary, so choose deliberately. Read and write are separate scopes per resource, and a token can never exceed what you ticked at creation time. A reasonable starting set for conversational CRM use covers contacts, conversations and conversation messages, opportunities, calendars and calendar events, custom fields, and locations — add write scopes only for the resources you actually intend Claude to change.

If you would rather not handle a token at all, HighLevel's v2 endpoint supports a one-click OAuth sign-in through its consent flow, which is the recommended path in HighLevel's multi-account article. Skip ahead and let Claude trigger the sign-in.

Both take a URL rather than a config file. Open Settings → Connectors → Add custom connector and paste the endpoint.

For HighLevel's server that is https://services.leadconnectorhq.com/mcp/anthropic/v2. With no credential in the URL, the connector prompts you through HighLevel's OAuth consent screen on first use, and you choose which sub-accounts to authorize there.

For Hylo, create a connector in the dashboard under Connect Claude and paste the URL it returns, which carries its credentials as query parameters:

https://api.hylo.pro/mcp/?client_id=YOUR_CLIENT_ID&client_secret=YOUR_CLIENT_SECRET

GoHighLevel connections already stored in the Hylo dashboard work immediately, with no setup inside the chat. Per-client setup notes live on the docs page.

Claude Code adds HTTP MCP servers from the command line, which writes the configuration for you:

# OAuth — Claude Code walks you through the consent flow
claude mcp add --transport http leadconnector \
  https://services.leadconnectorhq.com/mcp/anthropic/v2

# Or with a Private Integration Token instead
claude mcp add --transport http leadconnector \
  https://services.leadconnectorhq.com/mcp/ \
  --header "Authorization: Bearer pit-your-token" \
  --header "locationId: your-location-id"

The committed equivalent — what a project's .mcp.json looks like — is the same shape HighLevel publishes:

{
  "mcpServers": {
    "leadconnector": {
      "type": "http",
      "url": "https://services.leadconnectorhq.com/mcp/anthropic/v2"
    }
  }
}

Add Hylo alongside it the same way, using the API key from your dashboard:

claude mcp add hylo -e HYLO_API_KEY=hylo_sk_your_key_here -- npx -y hylo-mcp

Both read an mcpServers object — ~/.cursor/mcp.json for Cursor. HighLevel lists dedicated per-client v2 endpoints for Cursor and Windsurf as planned rather than available, so use the general endpoint with a Private Integration Token:

{
  "mcpServers": {
    "leadconnector": {
      "url": "https://services.leadconnectorhq.com/mcp/",
      "headers": {
        "Authorization": "Bearer pit-your-token",
        "locationId": "your-location-id"
      }
    },
    "hylo": {
      "command": "npx",
      "args": ["-y", "hylo-mcp"],
      "env": {
        "HYLO_API_KEY": "hylo_sk_your_key_here"
      }
    }
  }
}

The locationId header is optional on HighLevel's server — you can instead name the location in conversation and let the tool ask. Restart the editor after editing the file; neither client reloads MCP configuration in place.

Check the connection at the protocol level rather than by asking the assistant, because a model will happily describe a connection it does not have. A tools/list call returns exactly what your credential reaches:

curl -s -X POST https://services.leadconnectorhq.com/mcp/ \
  -H "Authorization: Bearer pit-your-token" \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

A healthy response is a server-sent event stream whose data: line carries a JSON tool list. Omitting the Accept header is the most common cause of a confusing failure here; sending no Authorization header at all returns HTTP 401 with {"error":"invalid_token"}.

Then ask Claude something read-only and specific — “how many contacts were created in the last seven days?” — so that a wrong answer is obvious. Save the first write for something reversible, such as adding a tag.

For the last row, the deprecated endpoints guide lists every endpoint GoHighLevel's specification currently marks as superseded, with the documented replacement for each.