GoHighLevel MCP Server: How to Connect Claude, ChatGPT and Cursor to HighLevel (2026)

The HighLevel MCP server lets AI assistants like Claude, ChatGPT and Cursor work directly with a GoHighLevel sub-account: look up contacts, check pipelines, read conversations and run actions, just by asking. This guide explains what it is, the two ways to connect, how to set it up in Claude and other tools, and how to use it without putting your CRM at risk.

1. What is the GoHighLevel MCP server?

MCP (Model Context Protocol) is an open standard that lets an AI assistant use outside tools. HighLevel runs an official MCP server on top of its API, so instead of writing code, you connect your AI assistant once and ask in plain English:

• "Which new leads came in this week that nobody has replied to?"
• "Show the opportunities stuck in the Quote Sent stage for more than 7 days."
• "Add the tag vip to Sam Lee and create a follow-up task for Friday."
• "Summarise my last conversation with this contact before I call them."

The assistant picks the right API operation, runs it with your permissions and reports back. It's the same idea as the GHL MCP servers people share on GitHub, but run and maintained by HighLevel.

2. Two ways to connect

HighLevel currently documents two versions of its MCP server:

All endpoints sit on https://services.leadconnectorhq.com.

3. How to connect GoHighLevel to Claude

Claude (web and desktop): add a custom connector in Claude's settings with the URL https://services.leadconnectorhq.com/mcp/anthropic/v2, then click Connect. A LeadConnector sign-in opens; log in, choose the sub-account and approve the permissions.

Claude Code (terminal): add it with one command, then run /mcp inside Claude Code to sign in:

claude mcp add --transport http leadconnector https://services.leadconnectorhq.com/mcp/anthropic/v2

ChatGPT and Codex use /mcp/openai/v2/ in the same way. Agency-wide access across many sub-accounts is still rolling out, so today you normally connect one sub-account at a time.

4. How to connect Cursor, Windsurf or a custom agent

For clients that send custom headers, use the original endpoint with a Private Integration token. Create the token in Settings → Private Integrations → Create new Integration, tick only the scopes the assistant needs, and copy it (it's shown once). Then add the server to your client's MCP config:

{
  "mcpServers": {
    "highlevel": {
      "type": "http",
      "url": "https://services.leadconnectorhq.com/mcp/",
      "headers": {
        "Authorization": "Bearer YOUR_PRIVATE_INTEGRATION_TOKEN",
        "locationId": "YOUR_SUB_ACCOUNT_ID"
      }
    }
  }
}

The tools available depend on the scopes you gave the token. If a tool is missing, the token usually lacks that scope. Custom agents built in n8n or LangGraph can connect the same way, or call the API directly.

5. MCP server vs API vs workflows: which to use?

In short: MCP is for people working with the CRM through AI. Customer-facing automation should still run as workflows or code, because an AI assistant can phrase or choose things slightly differently each time.

6. Security: how to use it safely

• Grant the fewest scopes possible. A reporting assistant needs read access, not the ability to send messages or take payments.
• Start with a test sub-account before connecting a live client account.
• Treat CRM data as untrusted input. Contact notes and incoming messages are written by outsiders. Don't let an assistant follow instructions found inside them, and review what it plans to do before it acts.
• Be careful with bulk actions and messaging. Ask for a preview before any change to many contacts, and never let an assistant text customers at scale. US texting still has to follow A2P 10DLC rules.
• Protect tokens. Keep Private Integration tokens out of shared config files and code repositories, and rotate them every 90 days.
• Mind client data rules. For UK and EU contacts, sending CRM data to an AI provider falls under GDPR, so check your agreements first.

HighLevel adds extra confirmation steps for sensitive operations in MCP v2, but the safest setup is still limited scopes plus a human checking anything that changes data.

7. Limitations to know

• Calls count toward the same API rate limits as other integrations on that sub-account.
• Agency-level access across many sub-accounts is still rolling out.
• The assistant only sees what the API exposes, so some settings in the GoHighLevel interface aren't available.
• Answers depend on how clearly you ask; for anything important, ask it to show the records it used.

Endpoints, tool lists and setup screens are changing quickly, so check HighLevel's current MCP documentation before relying on any guide, this one included.

Want AI connected to your GoHighLevel account?

I'm Arslan Mumtaz, a software engineer who connects GoHighLevel to AI: MCP setups for teams, Agent Studio agents, Conversation AI and custom agents built on the API. See AI & automation and pricing, or start with a $97 audit.