How to connect GoHighLevel to Claude or ChatGPT with the HighLevel MCP server

How do you connect GoHighLevel to Claude or ChatGPT?

Add HighLevel's official MCP server URL for your AI app as a custom connector, sign in through LeadConnector, pick the sub-account and approve the scopes. No Private Integration Token is needed. Claude, ChatGPT, Codex and Cursor each use their own URL.

This guide follows HighLevel's help article (last modified October 29, 2026) and its Marketplace developer docs, plus Anthropic's and OpenAI's connector docs, all read on October 30, 2026. It goes further than the short HighLevel section on our MCP server development page: the exact URL for each app, the steps, the scopes, the limits, and what to do when the connector does not work. AI app menus move often, so if a label below has changed, the order of steps usually has not.

New to the idea? Our short answer on how an MCP server differs from an API explains why Claude and ChatGPT need this layer at all.

Which GoHighLevel MCP URL should you use?

Use the URL made for your AI app: /mcp/anthropic/v2 for Claude, /mcp/openai/v2/ for ChatGPT and other OpenAI clients, /mcp/cursor/v2 for Cursor and /mcp/muse/v2/ for Muse, all on services.leadconnectorhq.com.

We compared the help article with the developer docs, and they match except for one line about Codex, flagged below.

How do you connect the HighLevel MCP server to Claude?

In Claude, open Customize, Connectors, choose Add custom connector, paste the Anthropic URL, then connect and complete the LeadConnector sign-in. Pick the sub-account, approve access, and turn the connector on in a new chat.

Claude on the web, Desktop or Cowork (Pro, Max and Free)

1. Open Customize, Connectors, select + Add, then Add custom connector. HighLevel's article calls this Settings, Connectors; Anthropic's current article uses Customize.
2. Name it (for example "HighLevel") and paste https://services.leadconnectorhq.com/mcp/anthropic/v2.
3. Keep the detected sign-in settings and select Add, then Connect.
4. Sign in to LeadConnector, choose the sub-account (or sub-accounts, if your agency has the multi-sub-account option), and approve the scopes.
5. Start a new chat, select +, then Connectors, and switch HighLevel on for that conversation.

Anthropic's article says custom connectors work on Free, Pro, Max, Team and Enterprise plans, and the Free plan allows one custom connector.

Claude Team or Enterprise

An owner adds the connector once under Organization settings, Connectors: select Add, hover over Custom, choose Web and paste the URL. On Team plans only Owners and Primary Owners can do this. Each member then opens Customize, Connectors, finds HighLevel and selects Connect, which runs their own LeadConnector sign-in. So each person's access follows the sub-accounts and scopes they approve.

Claude Code

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

A browser window opens for the LeadConnector sign-in the first time a tool is used. You can put the same server in a project's .mcp.json as a leadconnector entry with "type": "http" so everyone on the repository gets it.

How do you connect HighLevel to ChatGPT?

Turn on Developer mode in ChatGPT's Settings, open ChatGPT Plugins, add a new plugin or custom MCP server with the OpenAI URL, complete the LeadConnector sign-in, then add the connection from the tools menu in a new conversation.

1. Open Settings, Security and login and turn on Developer mode (HighLevel's step). On a work account, your workspace admin may need to allow it first: OpenAI says account and workspace policies apply to custom MCP servers.
2. Open ChatGPT Plugins, select the plus button and create a new plugin (OpenAI's docs call the option Add custom MCP server).
3. Enter a name and description. Under Connection, add https://services.leadconnectorhq.com/mcp/openai/v2/.
4. Configure sign-in, accept the risk warning, and complete the LeadConnector sign-in: choose sub-accounts and approve access.
5. Review the discovered tools. You should see four.
6. Start a new conversation and add the LeadConnector connection from the tools menu.

We could not open OpenAI's help article on Developer mode plans when we checked, so we do not list which ChatGPT plans include it. Check your own Settings screen.

How do you connect HighLevel to Codex or Cursor?

In Codex, run codex mcp add with the OpenAI URL, then codex mcp login. In Cursor, add a leadconnector entry with the Cursor URL to your mcp.json file. Both open the LeadConnector sign-in the first time.

Codex

codex mcp add leadconnector --url https://services.leadconnectorhq.com/mcp/openai/v2/
codex mcp login leadconnector

In the Codex IDE extension, open the gear menu, then MCP servers, Add server, choose Streamable HTTP, enter the URL and select Authenticate.

Cursor

{
  "mcpServers": {
    "leadconnector": {
      "url": "https://services.leadconnectorhq.com/mcp/cursor/v2"
    }
  }
}

Save it as .cursor/mcp.json in a project, or ~/.cursor/mcp.json for every project, and complete the sign-in when Cursor asks.

What can the HighLevel MCP server do once it is connected?

It exposes four tools that find, describe and run more than 550 read, create, update and delete operations across 38 areas of HighLevel, such as contacts, conversations, opportunities, calendars, payments and workflows, limited by the scopes you approved.

Four tools instead of 550 keeps the list the model chooses from short; it finds the right operation by searching. HighLevel suggests these prompts to test a new connection:

• "Find the last 5 contacts I added in HighLevel."
• "Find the contact with email jane@example.com, add the tag vip-2026", then create an opportunity for them in the Sales Pipeline.
• "List the contacts tagged vip-2026 in Downtown Clinic, then compare the count with Uptown Med Spa" (needs a multi-sub-account connection).

Use a test sub-account for the write prompts. For ideas on what AI can do inside HighLevel itself, without an outside app, see our GoHighLevel AI agents guide.

What permissions and scopes does the HighLevel MCP server ask for?

It asks you to choose sub-accounts and approve OAuth scopes during sign-in. Every operation lists the scopes it needs, and the server refuses anything outside your grant. Sensitive or permanent actions get extra confirmation before they run.

HighLevel does not publish one fixed scope list for the MCP server. The scopes for a given action appear when the assistant calls describe_operation, so you can ask "which scopes does this need?" before running it. Our advice, which goes beyond HighLevel's docs:

• Start read-only where you can. If the consent screen lets you leave out write scopes, do so for the first week and see what people actually ask for.
• Connect one sub-account first. Agencies should test on an internal sub-account before connecting client accounts. Our note on agency accounts versus sub-accounts explains the boundary.
• Watch Claude's research mode. Anthropic warns that during research Claude can call connector tools without asking for approval, and suggests disabling tools that write. With HighLevel, that means switching the connector off before you run research.
• Treat incoming text as untrusted. Contact notes, form answers and messages can contain instructions aimed at the model. Do not let the assistant send messages or delete records without a person checking.
• Revoke cleanly. HighLevel says you review or revoke access from your LeadConnector account. Also remove the connector in the AI app.

Patient data needs more care: read our guide to HIPAA-compliant automation with GoHighLevel before connecting a clinic's sub-account to any AI app.

What are the limits of the HighLevel MCP server?

Each request runs against one sub-account, agency-wide connections are still rolling out, and HighLevel does not publish separate MCP rate limits. Its API allows 100 requests per 10 seconds and 200,000 per day per app per sub-account, which makes bulk edits slow.

• One sub-account per request. Even with several sub-accounts connected, each operation targets one. If the account is unclear, the assistant calls list_locations and asks.
• Multi-sub-account is rolling out. If the sign-in only offers one sub-account, HighLevel says the option is not enabled for that app yet.
• No stated MCP rate limit or price. Neither HighLevel page we read gives one. The developer doc points to HighLevel's API rate limits page, which sets a burst limit of 100 requests per 10 seconds and 200,000 requests per day, counted per Marketplace app per sub-account or agency.

What those API limits would mean for bulk jobs

HighLevel does not say how MCP calls count against the API limits. If each record change is one API request (our assumption), the burst limit alone sets the minimum time below, before the time the model spends deciding each step:

The daily limit is rarely the problem. The real limit is that a chat assistant works through records one tool call at a time, and each call costs tokens and can stop partway through. For anything over a few dozen records, a HighLevel workflow or a script against the API is the better tool. Our GoHighLevel API integration work covers that route.

Why is the GoHighLevel MCP connector not working?

Most problems come from the wrong URL for your AI app, a sign-in that skipped a sub-account or scope, an old guide asking for a token, Developer mode being off in ChatGPT, or a connector that is added but not switched on in the chat.

We built this checklist from HighLevel's FAQ, Anthropic's and OpenAI's connector articles, and the MCP specification. Fixes that go beyond those sources are marked as ours.

Should you use HighLevel's official server, a community server, or a custom one?

Start with the official server: the docs we read mention no fee, it uses OAuth, and it covers most of the CRM. Choose a custom server for other systems in one tool, your own business rules, or a short list of safe actions.

Agencies running many client accounts have a further choice: connect each sub-account separately, wait for the multi-sub-account rollout, or build one server that applies the agency's own rules. Our automation work for GoHighLevel agencies covers that setup. Meta's agent took the official route; see our news note on Muse adding HighLevel as a connector.

More questions about the GoHighLevel MCP server

Does the GoHighLevel MCP server cost extra?

Neither HighLevel's help article nor its developer docs, read October 30, 2026, mention a fee for the MCP server. You still need a HighLevel account with the sub-accounts you connect, and an AI app plan that allows custom connectors. Claude's Free plan, for example, allows one custom connector, according to Anthropic's help article.

Can the AI create, update or delete records in GoHighLevel?

Yes, when the operation exists in the catalog and your sign-in approved the scope it needs. HighLevel says sensitive and irreversible operations go through extra confirmation and safety checks first. Test writes on a spare sub-account, and keep a person in the loop for anything that sends a message, charges money or deletes data.

Is there an official GoHighLevel MCP server on GitHub?

The official server is hosted by HighLevel at services.leadconnectorhq.com, so there is nothing to download or run. The GitHub projects that appear in searches for a GoHighLevel MCP server are community projects. Some ask for a Private Integration Token in a config file. Review their code and the scopes on that token before connecting a live account.

Can I use one connection for all my client sub-accounts?

Only where HighLevel has turned on the multi-sub-account option for the app, which its docs describe as still rolling out. Even then, each request targets a single sub-account and anything outside the installation is refused. Until then, agencies connect sub-accounts one at a time.

How do I disconnect GoHighLevel from Claude or ChatGPT?

Do both sides. Remove the connector in the AI app: in Claude, under Customize, Connectors, select the three dots and Remove. Then review or revoke the connection's access from your LeadConnector account, which HighLevel says is where grants are managed. Removing only the connector can leave the grant active.

Is it safe to connect GoHighLevel to ChatGPT or Claude?

It can be, with care. The official server uses OAuth, limits each connection to the sub-accounts and scopes you approve, and asks for confirmation on risky actions. The remaining risks sit with people and data: contact notes can contain prompt injection, and research modes can call tools without asking. Start read-only and keep writes under review.

Does the HighLevel MCP server work with n8n or my own agent?

HighLevel's docs list Claude, ChatGPT, Codex, Cursor, Muse, Grok Bot and other OpenAI-compatible clients. A server-side agent such as an n8n workflow cannot easily complete a browser sign-in for each user, so for back-office automation we usually call HighLevel's API directly. Our GoHighLevel automation page covers those builds.

Can you set this up for us?

Yes. On a free strategy call we check whether the official server covers what your team wants, connect and test it on your accounts, and tell you if a custom server is worth building. A small build of two or three automations is about $1,000 at the time of writing; larger work gets a fixed quote. Book the free call here.

Sources and further reading

• HighLevel, How to set up and use the HighLevel MCP Server (last modified October 29, 2026: URLs, steps per app, FAQ, test prompts), checked October 30, 2026.
• HighLevel, LeadConnector MCP Server developer docs (endpoints including the Codex path, tools, 550+ operations, multi-sub-account), checked October 30, 2026.
• HighLevel, API rate limits (100 per 10 seconds, 200,000 per day, per app per resource), checked October 30, 2026.
• Anthropic, Getting started with custom connectors using remote MCP (plans, menu paths, research mode warning, removal), checked October 30, 2026.
• OpenAI, Connect your MCP server to ChatGPT (Add custom MCP server, workspace policies, refresh), checked October 30, 2026.
• Model Context Protocol, Tools, specification 2026-07-28 (human in the loop, untrusted annotations), checked October 30, 2026.