Does GoHighLevel have an MCP server? Connecting Claude and other AI assistants

Yes: GoHighLevel runs an official MCP server

Yes. GoHighLevel has an official MCP server, so a GoHighLevel Claude integration exists without any third-party connector. MCP, the Model Context Protocol, is a standard way for an AI assistant to call tools in another system. HighLevel hosts the server itself on its LeadConnector infrastructure, the same place its public API runs.

HighLevel's documentation names Claude.ai, Claude Code, Claude Cowork, ChatGPT and Codex as supported clients on the current server. An older setup guide also names Cursor, Windsurf and the OpenAI Playground. What follows is drawn from those documents as read on 3 October 2026. We have not connected an account ourselves.

What the MCP server exposes to an AI assistant

HighLevel documents two versions of the server, and they look quite different.

On the current server the four tools list your sub-accounts, search for an operation by intent, describe what it needs and then execute it. The operations include reading, creating, updating and deleting records across contacts, conversations, opportunities, calendars, payments, invoices, forms, funnels, workflows and the phone system.

HighLevel's numbers are not consistent. Its Claude help article, last changed on 6 August 2026, says 625 operations across 40 product areas, while the developer docs say more than 550 across 38. Read either figure as approximate.

How to connect Claude to GoHighLevel

HighLevel's steps for Claude.ai are:

1. In Claude, open Settings and then Connectors.
2. Choose Add custom connector.
3. Paste the Claude server address shown in the table above, with https in front.
4. Click connect and sign in to HighLevel when the consent window opens.
5. Pick the sub-account, or several if you are an agency, and approve the scopes.
6. Start a new chat and ask for something small, such as listing recent contacts, to confirm it works.

For Claude Code the help article gives a single command that adds the same address as an HTTP server named leadconnector. ChatGPT and Codex use a separate address ending in /mcp/openai/v2/ with the same OAuth sign-in.

The token method is still documented for clients that cannot do OAuth. You create a Private Integration Token under Settings, then Private Integrations, choose its scopes, and add the token and your location ID as headers in the client.

Limits HighLevel documents for MCP

• One sub-account per request. Agencies can authorize several sub-accounts in one connection and switch by naming the account, but each operation runs against a single one.
• Scopes decide what works. The assistant only sees operations that the OAuth scopes or token scopes allow.
• Confirmation on risky actions. The developer docs say sensitive and irreversible operations require an extra confirmation step.
• No published MCP rate limit. None of the three MCP pages states one. The API limits of 100 requests per 10 seconds and 200,000 per day are published for the API, not for MCP.
• No published plan or price. None of the MCP pages names a required plan or a fee. Messages, calls and AI features an assistant triggers are still billed as normal usage.
• Conflicting documents. The older guide, last changed on 9 July 2026, still says OAuth is planned for later, while the newer pages describe it as live.

Who should connect an AI assistant, and who should not

The MCP server suits an owner or agency that already works in Claude or ChatGPT and wants to ask questions of the CRM in plain language, such as which leads from last week have no reply, or to make small updates without opening the app.

It is not a replacement for automation. An assistant only acts when someone asks it to. A text that must go out every time a call is missed belongs in a workflow, as described in how missed call text back works. And because the server can delete and update records, give it narrow scopes first. A business that handles health information should check the HIPAA terms on is GoHighLevel legit before sending CRM data to any outside assistant.

If you need a fixed, repeatable sync between systems, use the API or a connector from our GoHighLevel integrations guide instead.