How to connect Claude to GoHighLevel with MCP and Claude Code (2026)

Members can't add custom connectors themselves. An Owner or Primary Owner goes to Organization settings > Connectors > Add > Custom > Web and adds the same URL. Then each person goes to Customize > Connectors, finds it and clicks Connect, which runs their own HighLevel sign-in, so everyone only gets the sub-accounts and permissions they approve.

Run this in your terminal: claude mcp add --transport http leadconnector https://services.leadconnectorhq.com/mcp/anthropic/v2. Add --scope user if you want it in every project, otherwise it only loads in the folder you ran it in.

Then open Claude Code and type /mcp, pick leadconnector and finish the browser sign-in (or run claude mcp login leadconnector from the shell). claude mcp list should show it as Connected. If it says Needs authentication, you skipped the sign-in.

You can also put it in a .mcp.json file at the project root with "type": "http" and the URL, which is handy if your team shares a repo. Everyone still signs in with their own HighLevel login. Don't put an Authorization header in there: if a header gets rejected, Claude Code won't fall back to the OAuth sign-in.

To disconnect properly, run claude mcp logout leadconnector before claude mcp remove leadconnector. Removing it on its own only deletes the local copy of the login.

The server gives Claude four tools: list_locations, search_operations, describe_operation and execute_operation. Behind them sit more than 550 operations across 38 areas of HighLevel. Claude searches for the operation it needs, reads its inputs and safety notes, then runs it, which is why you see four tools in Claude and not hundreds.

What it can actually do depends on the permissions you approved. If you're not sure, ask Claude to use search_operations for the thing you want. If it's not in the catalog, it can't do it.

Yes, but only in private beta right now. HighLevel added workflows to the MCP server on October 28, 2026, after a request for exactly this ("Claude > GHL workflow automations (write)") sat at 399 votes on their ideas board. Claude can find a workflow, read its triggers and steps, build one from a description, publish it, delete it, read the execution logs and count enrollments and failures per step.

It only works in sub-accounts enabled for the Workflows API. Everywhere else Claude gets WORKFLOW_API_NOT_ENABLED_FOR_LOCATION, and you join the beta through HighLevel support or the form linked in their changelog. Back in June I said in my video it would soon be able to create automations, and here we are.

The rules to know before you let it touch a live workflow: new workflows are always saved as drafts, and publishing is a separate request. To change a published workflow it gets set back to draft first, which turns its triggers off and pauses everyone in it at their next step until it's published again. An update replaces the whole workflow, and if someone is editing it in the builder at the same time the update gets rejected. Deleting is permanent and stops everyone in it.

Logs cover up to 30 days per request and go back 60 days. They only show live runs, so contacts you pushed through with Test workflow won't appear. My favourite use is the boring one: ask which steps failed this week and for whom. For how workflows are built by hand, see GoHighLevel workflows.

Not for Claude. HighLevel's own FAQ says the Claude, ChatGPT and Cursor connections use OAuth, so you don't create or rotate a token. HighLevel also says OAuth gets a wider set of permissions than a token does.

You'd use a Private Integration token for the older universal endpoint (https://services.leadconnectorhq.com/mcp/), which works with any MCP client but has a smaller set of tools and one sub-account per connection, and for n8n, where HighLevel's example passes the token as Authorization: Bearer plus a locationId header. A lot of the guides ranking for this still only teach that older setup.

If you do need one: in the sub-account go to Settings > Private Integrations (if it's missing, check Labs), click Create new Integration, name it after what it's for, tick only the scopes that job needs, and copy the token straight away because HighLevel only shows it once. You can have up to 20 per sub-account and 20 at the agency level, so give each tool its own.

HighLevel recommends rotating tokens every 90 days. Rotate and expire this token later gives you 7 days where the old and new token both work. If a token leaks, use Rotate and expire this token now. You can edit a token's scopes later without generating a new one. In Claude Code, keep it out of a shared .mcp.json by using an environment variable like ${GHL_TOKEN} in the header.

A normal connection is tied to the one sub-account you pick at sign-in. Agency connections that cover several sub-accounts are rolling out: if your sign-in screen lets you choose more than one, you have it, and you just name the client in the prompt ("in Downtown Clinic, list contacts tagged vip"). Every request still runs in one sub-account, and Claude can't reach a sub-account you didn't select.

It works on a white-labelled account. Mine is Hexona Systems, and the Social Planner you see in the video is my white-label version. The sign-in screen is the LeadConnector one, so if a client connects their own Claude, that's the name they'll see. More on that setup in White label guide.

Connect a test sub-account first. Build a sandbox with fake contacts and a copy of your workflows, and let Claude loose there before it ever sees a client. I didn't even reconnect mine on camera because I didn't want it to mess up the one I already had.

Approve only the permissions you need. On the consent screen, if the job is reporting, leave the send and delete permissions off. For workflows, reading, changing, publishing, deleting and reading logs are each separate permissions, so you can give Claude read access and the logs without letting it publish.

Be careful with Always allow. Claude asks before it runs a tool, and you can pick allow once or always allow. Always allow on read tools is fine. On sending messages, deleting or anything with payments, approve each one. HighLevel adds its own confirmation on deletes and other irreversible actions too.

Use the official server. There are a bunch of community GoHighLevel MCP servers on GitHub, and some of their setup guides tell you to tick every read and write scope on your token. Anthropic's own docs warn that a custom connector isn't verified by them and a malicious server can hide instructions for Claude. Also skip the "hacks" that call HighLevel's internal endpoints to write workflows. They aren't part of the public API, so they can break without warning.

Revoke access when a project ends. Remove the connector in Claude, revoke it from your LeadConnector account, and delete any Private Integration token you made for it.

In my video I asked Claude to make a fun before and after Facebook post for a pet groomer. It made the image with another tool I have connected (Higgsfield), wrote the copy and scheduled it in the Social Planner. I refreshed HighLevel and it was there, at the time it said.

Then I asked for a report on recent contacts and unread messages. It came back with 54 contacts and 42 unread messages across 15 conversations in my demo account. I told it to text one of the contacts hello and it showed up in Conversations a few seconds later.

One funny thing: it told me the image tool's login had expired, which wasn't true. All I said was "it's connected, just run it" and it carried on. Sometimes you have to tell it straight up that it can do something.

Other jobs that work well today: tagging and cleaning up contacts, spotting stale deals in a pipeline, moving opportunities, drafting email templates and blog posts, auditing an inherited account, and bulk changes to custom values. With Cowork you can put the content ones on a schedule, like two or three posts every morning.

HighLevel launched an official app in ChatGPT on February 27, 2026. In ChatGPT go to Settings > Apps, search for HighLevel and connect it. HighLevel pitches it for questions about your pipeline, stalled deals and campaigns.

If you want the full MCP catalog in ChatGPT, HighLevel's current steps are: turn on Developer mode under Settings > Security and login, open ChatGPT Plugins, create a new plugin and add https://services.leadconnectorhq.com/mcp/openai/v2/ as the connection. Codex uses the same URL. Your ChatGPT plan and workspace settings decide if this option is there for you.