GoHighLevel MCP: How to Connect Claude or ChatGPT to Your Account

Yes, GoHighLevel has an official MCP server, and it connects Claude, ChatGPT, and Codex to your CRM through OAuth. HighLevel's MCP documentation lists more than 550 active operations across about 40 areas, from contacts and pipelines to calendars and invoices.

Start read-only on a test sub-account. Every connection works on one sub-account at a time, and the assistant can only do what the scopes you approve allow.

• Claude: services.leadconnectorhq.com/mcp/anthropic/v2, with OAuth.
• ChatGPT and Codex: services.leadconnectorhq.com/mcp/openai/v2/, with OAuth.
• Older method: services.leadconnectorhq.com/mcp/, with a Private Integration Token and narrower scopes.
• Cost: HighLevel's docs list no MCP fee. A text or email the assistant sends bills like any other from your wallet.
• Best first prompt: ask for your last 5 contacts. That only reads data.

Which GoHighLevel MCP Endpoint Should You Use?

HighLevel runs 4 MCP endpoints, and the right one depends on your AI client. Claude, ChatGPT, and Muse each get their own OAuth endpoint with a separate grant. Any other HTTP-based client, such as Cursor or Windsurf, uses the older endpoint with a Private Integration Token.

All four live at services.leadconnectorhq.com. OAuth is the recommended route because there is no token to store or rotate, and the scope list is wider.

Expand

What the HighLevel MCP Server Can Do

HighLevel's MCP server exposes more than 550 active operations across about 40 areas, but only a handful of tools. The assistant calls search_operations to find an action, describe_operation to read its inputs, and execute_operation to run it. list_locations picks the sub-account.

• Contacts and conversations: find, create, update, and tag contacts, add notes and tasks, read and send messages, and pull recordings and transcriptions.
• Sales: pipelines, opportunities, status changes, invoices, estimates, payments, and subscriptions.
• Scheduling: calendars, appointments, free and blocked slots, and bookings.
• Marketing: email templates and campaigns, forms, surveys, funnels, blogs, and the Social Planner.
• AI and phone: Voice AI agents and call logs, number pools, and phone number purchasing.
• Content and data: knowledge bases, media files, custom objects and fields, and brand boards.
• Account setup: sub-account lookup, tags, custom values, users, and businesses.
• Money and ads: wallet transactions, supported Facebook, Google, and LinkedIn ad operations, and affiliate commissions.

What the assistant can really do depends on the scopes you approve. Ask it to run search_operations at any time and it will list exactly what your grant allows. Our GoHighLevel CRM guide explains what the contact and pipeline objects behind these operations do.

How to Connect Claude to GoHighLevel

Claude connects to HighLevel through one URL and an OAuth sign-in. You pick the sub-account and approve scopes on HighLevel's consent screen, with no token to copy. HighLevel documents steps for Claude.ai, Claude Code, and Claude Cowork. All three use the same endpoint ending in /mcp/anthropic/v2.

1. In Claude.ai, open Settings, then Connectors, then Add custom connector.
2. Set the server URL to https://services.leadconnectorhq.com/mcp/anthropic/v2.
3. Click Connect and sign in to HighLevel.
4. Pick the sub-account to expose, review the scopes, and approve.
5. Start a new chat and ask: "Find the last 5 contacts I added in HighLevel."

In Claude Code, add the server with one command. The first time you use it, Claude Code opens a browser for the HighLevel sign-in. Run claude mcp list afterward to confirm it is connected.

You can also commit the same setting to a project file. Add this to .mcp.json at your project root instead of running the command.

In Claude Cowork, open the connectors settings, choose Add custom connector, and use the same URL.

How to Connect ChatGPT or Codex to GoHighLevel

ChatGPT and Codex use a different endpoint from Claude, ending in /mcp/openai/v2/. ChatGPT needs Developer mode switched on before it accepts a custom MCP connection. Codex adds the server with 2 terminal commands. Each client keeps its own scopes, so approving one does not grant the other.

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 https://services.leadconnectorhq.com/mcp/openai/v2/ as the MCP server URL.
4. Complete the HighLevel sign-in, pick the sub-account, and approve the scopes.
5. Start a new conversation and add the connection from the tools menu.

For Codex, add the server and then log in. The second command starts the OAuth flow.

When to Use the Older Private Integration Token Method

The older endpoint at /mcp/ still works with any HTTP-based MCP client, including Cursor, Windsurf, and OpenAI Playground. It authenticates with a Private Integration Token plus the sub-account's location ID. HighLevel says it stays supported, but it exposes a narrower scope list than the 3 OAuth endpoints.

1. Open the sub-account, go to Settings, then Private Integrations.
2. Click Create New Integration and choose only the scopes you need.
3. Copy the generated token and store it like a password.
4. Add the server to your client with the token and location ID in the headers.

HighLevel's setup guide lists the token method's 36 example tools. HighLevel also says its MCP server works with n8n without a custom plugin, using this same configuration.

Which Scopes to Grant: Start Read-Only

Scopes decide what the assistant can touch, so grant them in 3 stages. Start with view scopes only. Add edit scopes for contacts and opportunities once the reads look right. Add messaging last, because a text or email reaches a real person and bills your wallet.

The scope names above come from HighLevel's recommended list for the token method. The staging is my own advice, built on HighLevel's best practices of least privilege and testing before production.

Operator scenario: Say you manage 12 client sub-accounts. Connect one test sub-account with view scopes first. Ask for last week's unanswered conversations. If the summary matches the inbox, repeat on one live client before you add any edit scope.

Prompts to Try First

Good first prompts read data and ask for a summary, not an action. The first 4 prompts below work with view scopes only. The last one writes data and belongs on a test sub-account. Replace the bracketed parts with your own pipeline, tag, or date range.

The agency prompt works only if your install can include several sub-accounts. Otherwise connect each sub-account separately.

Limits and Risks to Know Before You Connect

The server has 4 limits that shape what you can automate. HighLevel adds confirmation to irreversible operations, but a prompt that is too broad can still change many records. Plan around the limits below before you give an assistant a bulk job.

• One sub-account per request. Even an agency connection cannot run a single action across several sub-accounts.
• Agency installs are still rolling out. If your install offers only one sub-account, the multi-account option is not enabled for your app yet.
• Scopes cap everything. If a scope is missing, the matching operation does not appear for the assistant.
• API rate limits. HighLevel's published API limits are 100 requests per 10 seconds and 200,000 a day, counted per app and per sub-account. MCP operations run against that API, so a bulk read can hit the burst limit.

Two more risks sit outside HighLevel's control.

• Inbound content is untrusted. Contacts can type anything into a text, form, or chat. If the assistant reads that content while holding edit scopes, a message could try to steer it. Keep edit scopes off in chats that read inbound messages.
• Tokens leak easily. A Private Integration Token pasted into a shared prompt, screenshot, or public repo works for anyone who finds it. HighLevel says to store tokens securely and rotate them. OAuth avoids the problem because there is nothing to paste.

Messages and calls the assistant triggers use the same wallet-billed rates as everything else. Our GoHighLevel usage rate card lists them.

HighLevel as the MCP Client: Connect Outside Tools to Your AI Agents

HighLevel can also connect out to other MCP servers. Workflow AI agents and Voice AI agents can call their tools. Examples include Exa, Tavily, Browserbase, and your own server. HighLevel supports 5 authentication types and two transport options. It does not host third-party servers, and their fees are separate from your plan.

• Workflows. In a workflow's AI Agent action, choose Add Tools, open the MCP tab, add a connection, and press Test Connection. HighLevel lists the discovered tools, and you enable only the ones the agent needs.
• Voice AI. MCP support for Voice AI agents is a Labs feature that you switch on in agency settings. You add the server, assign tools per agent, and describe when each tool should run.
• If the server is down. A Voice AI agent keeps talking but cannot run actions that depend on the missing tool.

Who Should Use the GoHighLevel MCP Server, and Who Should Skip It

The MCP server suits operators who already work in Claude or ChatGPT and want answers from their CRM without clicking through dashboards. It is a poor fit for 3 groups. Some people need bulk edits in one request. Others want no-code automation. Others will not give an AI write access to customer records.

• Use the MCP server if: you review pipelines or conversations often, manage several sub-accounts, or build agents that need CRM context. See what GoHighLevel includes before you decide.
• Use HighLevel workflows instead if: the task repeats on a trigger, such as a new lead or a missed call. Workflows run without a chat.
• Skip it for now if: you cannot test on a separate sub-account.

You need a HighLevel account before you can connect anything. Our free trial guide shows how to get 30 days instead of 14.