Does GoHighLevel Have an MCP Server? Connect Claude and ChatGPT

Yes. HighLevel has an official MCP (Model Context Protocol) server, documented under the name LeadConnector MCP server. It lets an AI assistant such as Claude or ChatGPT find and change records in your HighLevel sub-account, after you sign in with OAuth and approve the scopes. Checked on October 15, 2026 against HighLevel’s marketplace developer docs. The docs page carries no publish date, so treat the details below as current as of that check and re-read the page before you build on it.

This page answers the problem most people searching “gohighlevel mcp” actually have: can I point Claude or ChatGPT at my account, how do I do it, and what could go wrong.

What is the GoHighLevel MCP server?

MCP is an open protocol that gives AI assistants a standard way to work with an app’s data. HighLevel’s LeadConnector MCP docs describe the server as a secure connection between MCP-compatible assistants and your CRM. Based on the scopes you approve, an assistant can find records and run supported read, create, update and delete actions. “LeadConnector” is HighLevel’s own app name, covered in what is LeadConnector.

HighLevel’s example prompt shows the idea: “Find the contact with email jane@example.com, add the tag vip-2026, and create a new opportunity in the Sales Pipeline worth $5,000.” Behind that one request the assistant searches for the right operations, inspects their inputs and scopes, then runs them one at a time.

The launch was announced in a HighLevel press release dated July 28, 2025, which described 21 tools and a token-based setup. The current docs describe a different, larger design (four tools in front of a catalog, OAuth). If you find a guide that mentions 21 tools or a Private Integration Token, it likely describes that earlier version. No dated entry for the MCP server appears in the HighLevel changelog, so we cannot date the switch to OAuth.

Can Claude connect to GoHighLevel?

Yes, and HighLevel documents three Claude paths. All use the same endpoint: https://services.leadconnectorhq.com/mcp/anthropic/v2.

Client	How HighLevel’s docs say to connect
Claude.ai	Settings, Connectors, Add custom connector. Enter the URL, click Connect, then sign in, pick the sub-account(s) and approve. Start a new chat.
Claude Code (CLI)	Run claude mcp add --transport http leadconnector https://services.leadconnectorhq.com/mcp/anthropic/v2, or add an mcpServers entry with "type": "http" and the URL to .mcp.json. A browser opens for authorization on first use. Check with claude mcp list.
Claude Cowork	Open the connectors or integrations settings, Add custom connector, enter the URL and complete the sign-in.

Claude Desktop is not named in HighLevel’s docs. Do not assume a pasted claude_desktop_config.json entry will work: the docs describe remote OAuth connectors, and some third-party guides that cover Desktop use community servers instead, which are not HighLevel’s. To confirm the connection, HighLevel suggests asking Claude to find the last 5 contacts you added.

Can ChatGPT connect to GoHighLevel?

Yes. OpenAI-compatible clients use a different endpoint: https://services.leadconnectorhq.com/mcp/openai/v2/ (note the trailing slash). HighLevel’s steps for ChatGPT:

1. In ChatGPT, open Settings, then Security and login, and turn on Developer mode.
2. Open ChatGPT Plugins, select the plus button and create a new plugin.
3. Enter a name and description. Under Connection, add the OpenAI endpoint as the MCP server URL and create the connection.
4. Complete the LeadConnector OAuth flow: sign in, pick sub-account(s), approve.
5. Start a new conversation and add the LeadConnector connection from the tools menu.

For Codex, run codex mcp add leadconnector --url https://services.leadconnectorhq.com/mcp/openai/v2/ and then codex mcp login leadconnector. In the Codex IDE extension, use the gear menu, MCP servers, Add server, choose Streamable HTTP and enter the same URL. HighLevel says other compatible OpenAI clients can use the endpoint as a remote Streamable HTTP MCP server. Muse has its own official LeadConnector app and endpoint (/mcp/muse/v2/).

ChatGPT’s menu names and plan requirements are OpenAI’s and change often, so check OpenAI’s current help if a step does not match your screen.

What does the GoHighLevel MCP server let an AI do?

The server does not list 550 tools. It shows four, and the assistant uses them to discover the rest:

Tool	What it does (per the docs)
list_locations	Lists the sub-accounts the connection can use, or returns the one bound sub-account.
search_operations	Finds operations by intent, including record searches and read, create, update or delete actions.
describe_operation	Shows an operation’s inputs, example payload, required scopes and safety information before it runs.
execute_operation	Runs the operation, subject to scopes, permissions, tenant boundaries and built-in safety checks.

Behind them sits a catalog of more than 550 operations across 38 domains. The docs list areas including contacts, conversations and messages, opportunities and pipelines, calendars and appointments, payments, invoices and estimates, products and store, social planner, blogs, emails, forms and surveys, funnels, Voice AI, phone system, knowledge base, media library, custom objects and fields, locations and CRM settings, users, wallet, and ads and affiliates. Workflows are limited: the docs say workflow lookup is available, with contact enrollment and removal done through contact operations. The docs do not say an assistant can build or edit a workflow, so do not plan on that. Ask search_operations for the exact list your connection can reach.

Can HighLevel’s AI connect to other MCP servers too?

Yes, that is the opposite direction and a different feature. HighLevel’s help article MCP connectors for Managed Agents says Managed Agents (in Agent Studio) and Ask AI can connect to third-party MCP apps, with a catalog that includes HubSpot, Notion, Make, Calendly, Monday.com, Stripe, Zapier and more, and that custom MCP connectors are supported too. You pick which tools each agent may use. That is HighLevel’s AI acting on other apps. This page covers the reverse: your own assistant, such as Claude, acting on HighLevel.

How does authentication work?

The current docs say connections use OAuth. Your client opens the LeadConnector sign-in and consent screen. You choose the sub-account (location) or sub-accounts and approve scopes. The integration can perform only what those scopes allow, and you can review or revoke access from your LeadConnector account. No manual token creation or rotation is needed.

Every request targets a single sub-account. Agencies can install once, pick which sub-accounts to include and then name the sub-account in each request, for example “in Downtown Clinic, list contacts tagged vip-2026”. A request for a sub-account you did not include is refused. The docs add a caveat: individual single-sub-account connections work today, and agency-wide connections are rolling out through a multi-sub-account (bulk) install option, so yours may not show it yet. For how sub-accounts work in general, see GoHighLevel sub-accounts explained.

What are the limits of the GoHighLevel MCP server?

HighLevel’s MCP page does not publish a rate limit, a usage fee or a plan requirement. We checked the page for each and found none. What the docs do state:

Limit or rule	What the docs say
Rate limits	Not stated on the MCP page.
Sub-account scope	One sub-account per request. Agency multi-account install is rolling out.
Operations	Depend on the scopes and permissions you approve.
Safety	Sensitive and irreversible operations get additional confirmation.
Availability by client	Endpoints exist for Claude, OpenAI-compatible clients and Muse.

The numbers that circulate, 100 requests per 10 seconds and 200,000 per day, come from the v2 API and are covered in our GoHighLevel API integration guide. We could not confirm from the MCP docs that MCP traffic counts against those caps, so treat them as a reasonable planning ceiling, not a documented MCP limit. Messages sent through MCP are still messages: SMS, email and voice usage rates apply as usual, per HighLevel usage rates.

Is it safe to connect an AI to my GoHighLevel account?

It is as safe as the scopes and the test habits you bring. The catalog includes delete and update actions, payments and invoices, and messaging, and an assistant can misread a request. HighLevel’s own safeguards are the scope grant, the sub-account choice and extra confirmation on sensitive or irreversible operations. Add your own:

1. Start in a sandbox sub-account. Connect to a test sub-account with fake contacts first and confirm what the assistant can reach with search_operations.
2. Approve the narrowest scopes that fit the job. If you only need to read contacts, do not grant delete or payment scopes.
3. Ask for a plan before an action that changes many records, such as bulk tagging, and read the describe_operation output.
4. Be careful with anything that sends. Messages, emails and Voice AI calls reach real people and carry consent and TCPA obligations.
5. Use the right user. The connection acts through the account you sign in with, so use a user whose role and permissions match the job.
6. Review and revoke. HighLevel says you can review or revoke access from your LeadConnector account. Do it when a project ends.

MCP, Zapier or Make, n8n, or the API: which should I use?

Pick by who starts the job and how often it repeats.

If you need	Best fit	Why
A person to ask for one-off work in plain language (“tag these, summarize that pipeline”)	MCP	The assistant finds and runs the operations. No build step.
The same action every time an event happens (form submitted, stage changed)	HighLevel workflows, then Zapier or Make for other apps	Deterministic triggers do not depend on a model reading a prompt.
Connections to many other apps with no code	Zapier or Make	Broad app coverage. Cost is per task or operation. See Zapier vs native integrations.
Self-hosted or branching logic you control	n8n or a webhook-driven build	You own the logic. HighLevel documents webhooks and the API. We did not review n8n for this page.
High volume, custom code or a product feature	The v2 API and webhooks	Documented rate limits and headers. See the API guide.
An agent that talks to your leads	Conversation AI or Voice AI inside HighLevel	Built for customer-facing conversations. See the Conversation AI guide.

A rule of thumb: use MCP where a human is in the loop, and use workflows or the API where nobody is watching. An assistant can phrase the same request differently on a second day. A workflow cannot.

Need MCP set up safely for your agency?

Scopes, sub-account selection and test plans take a few hours to get right and a bad grant is expensive. aibrevo works with its partner company autoesta, whose team delivers GoHighLevel and AI automation builds. See our GoHighLevel implementation service or contact us to scope it.

Sources

• HighLevel, LeadConnector MCP Server (marketplace developer docs): endpoints, OAuth, four tools, 550+ operations across 38 domains, client setup steps, agency multi-sub-account note, safety statement. Page undated, checked October 15, 2026.
• HighLevel Support, MCP connectors for Managed Agents: the reverse direction, third-party MCP apps inside Managed Agents and Ask AI. Undated, checked October 15, 2026.
• HighLevel, Highlevel MCP feature board: user feature requests only, no release dates. Checked October 15, 2026.
• HighLevel, Changelog searched for MCP: no dated MCP server entry found. One mobile app entry mentions MCP actions in Ask AI and is undated in the text we read. Checked October 15, 2026.
• HighLevel via ACCESS Newswire, HighLevel Expands Its AI Leadership With MCP Server for Action-Ready Automation, July 28, 2025: 21 tools at launch, token-based setup. Earlier version, checked October 15, 2026.
• aibrevo, GoHighLevel API integration guide: v2 API rate limits (100 per 10 seconds, 200,000 per day), cited from HighLevel’s API docs on that page.