Connect Claude to GoHighLevel with HighLevel's Official MCP Server: Setup, Safe Testing and Workflows (Private Beta)

Short answer: HighLevel now runs its own MCP server, the LeadConnector MCP server. To connect Claude, add a custom connector in Claude with the URL https://services.leadconnectorhq.com/mcp/anthropic/v2, click Connect, sign in to HighLevel, choose a sub-account and approve the permissions. There is no API key to paste. Then turn the connector on in a chat and start with read-only questions such as "Find the last 5 contacts I added".

Quick facts

What the HighLevel MCP server does

MCP (Model Context Protocol) is an open standard that lets an AI assistant use tools in another app. HighLevel's MCP server gives Claude access to your HighLevel data through the same permissions you approve when you connect. If you want the background on MCP itself, read What Is Model Context Protocol (MCP)?

HighLevel's developer docs list what you can work with, including:

• Contacts, conversations, opportunities and pipelines
• Calendars and appointments
• Social Planner, blogs, emails and forms
• Voice AI call logs
• Workflows (private beta, covered below)

Claude does not see hundreds of separate tools. HighLevel's docs describe a small set of tools: search_operations finds the operation it needs, describe_operation reads how that operation works, and execute_operation runs it. The developer docs also list list_locations, which shows the sub-accounts the connection can use. Behind them sit hundreds of operations, filtered by the permissions you approve. HighLevel's pages give different totals, so check the developer docs for the current list.

Claude can only do what the permissions you approved allow, and HighLevel says sensitive and irreversible operations are gated with extra confirmation and safety checks before they run.

Step 1: Pick the right link

Use the Claude link, not the older one

HighLevel gives each AI client its own server URL. There are two links you will see in guides, and they work differently:

HighLevel also publishes separate links for ChatGPT (/mcp/openai/v2/) and Cursor (/mcp/cursor/v2). Only copy server links from HighLevel's official docs. Anthropic warns that custom connectors connect Claude to services it has not verified.

Step 2: Add HighLevel as a custom connector in Claude

Open Connectors

In Claude on the web or Claude Desktop, go to Customize > Connectors. HighLevel's docs call the same place Settings > Connectors.

Add the custom connector

1. Click + Add, then Add custom connector.
2. Name it something clear, such as HighLevel.
3. Paste https://services.leadconnectorhq.com/mcp/anthropic/v2 as the URL and click Continue.
4. Claude checks the server and shows the authentication settings it detected. Review them and continue.
5. Leave the OAuth client on Use Claude's published identity (Recommended). You do not need request headers for this link, because sign-in happens through OAuth.
6. Click Add. HighLevel now appears in your connector list with a Connect button.

On Claude Team and Enterprise plans, Anthropic says an Owner adds custom connectors in Organization settings > Connectors, and each member then connects with their own sign-in. For Claude Code, HighLevel's docs give this command, and claude mcp list confirms it was added:

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

Step 3: Connect, sign in and approve access

Connect and sign in to HighLevel

Click Connect next to the new connector. A HighLevel sign-in page opens. Sign in with your LeadConnector (HighLevel) account.

Choose the sub-account and approve the permissions

You will pick a sub-account and see a list of permissions. HighLevel's own pages show these screens in different orders, so follow what you see. Pick the sub-account Claude should work in: a demo or test sub-account is the safest place to start. Then read the consent screen before you approve it: it lists what Claude will be allowed to do.

Our advice, not a HighLevel rule, is to start by approving viewing only, for example contacts, conversations, opportunities and calendars, and add edit permissions later once you are happy with the answers. When you approve, the connector shows as connected.

HighLevel says each request targets one sub-account. Its developer docs and changelog say agency connections that cover several sub-accounts are rolling out, so check the current docs if you manage many locations.

Step 4: Turn it on in a chat and run read-only prompts

Turn on the connector

Start a new chat. Click the + button in the message box, open Connectors and make sure the HighLevel toggle is on.

Ask read-only questions first

HighLevel's docs suggest this as the first test: "Find the last 5 contacts I added". Other useful read-only prompts:

When Claude wants to use a tool, it can ask for your approval. Anthropic's docs say Allow once approves that call, and Always allow stops the prompt for that tool. While you test, Allow once keeps you in control. You can change this later under the connector's Tool permissions, where each tool can be set to Always allow, Needs approval or Blocked.

Check the first few answers against the CRM. AI answers can be wrong, and a quick comparison tells you whether Claude used the right sub-account and filters.

Step 5: Run one safe write test

Tag a test contact

Once read-only answers look right, and only if you approved edit permissions for contacts, try one small change on a test contact, not a real lead. Create a contact called "MCP Test" first, then ask:

Find the contact "MCP Test" and add the tag mcp-test and a note that says "connected from Claude".

Open the contact in HighLevel and check both the tag and the note are there. If something looks wrong, remove the tag by hand and check which permissions you approved.

Using another MCP tool: the Private Integration Token route

If your MCP client does not support HighLevel's OAuth sign-in, HighLevel's help docs describe the older link, https://services.leadconnectorhq.com/mcp/, with a Private Integration Token (PIT):

1. In the sub-account, go to Settings > Private Integrations.
2. Click Create New Integration and give it a recognisable name and description.
3. Choose scopes. HighLevel's help article lists View and Edit scopes as recommended and says to grant only what your workflow needs. For read-only use, our advice is to pick only View scopes, such as View Contacts, View Conversations, View Opportunities, View Calendars and View Calendar Events.
4. Click Create Integration and copy the token.

Your MCP client then sends the token and the sub-account's location ID with every request, as headers:

"Authorization": "Bearer <your-token>",
"locationId": "<your-location-id>"

HighLevel says to treat a Private Integration Token like a password. Never paste it into a chat or a shared document. If it leaks, delete it in Private Integrations and create a new one. For a full walkthrough of scopes, see the GoHighLevel Private Integration Token setup guide.

Workflows in the MCP server Private beta

On 28 October 2026 HighLevel's changelog announced workflows in the MCP server, as a private beta. Workflow requests use the Workflows API, which HighLevel says is enabled for selected sub-accounts only. In other sub-accounts, requests are refused with the error WORKFLOW_API_NOT_ENABLED_FOR_LOCATION, and HighLevel says to contact support to join the beta.

In sub-accounts that have it, HighLevel's help article says an assistant can:

• Find workflows by name, status, folder or date, and explain one by reading its triggers, steps, branches and settings
• Read execution logs: who entered, what happened at each step, and which steps failed, with the error
• Count enrollments, completions and failures per step, by hour, day or week
• Build a workflow from a description and save it as a draft
• Update, publish, set back to draft or delete a workflow, if you approved those permissions

Example prompts based on HighLevel's help article:

Which steps in my New Lead Nurture workflow failed this week, and for which contacts?
How many contacts completed my Webinar Follow-up workflow each week this month?
Draft a workflow that tags new contacts webinar-lead and sends a welcome SMS.

Rules worth knowing before you try it:

• New workflows are always created as drafts. Publishing is a separate request.
• To change a published workflow, it is set to draft first. While it is a draft, its triggers are off and contacts pause at their next step until it is published again.
• An update replaces the whole workflow. If someone edits the same workflow in the builder at the same time, the update is rejected.
• Deleting a workflow is permanent and stops everyone currently in it.
• Logs cover up to 30 days per request and go back 60 days. They show live runs, not runs started with Test workflow, and include contact names and emails.
• Read, change, publish, delete and logs are separate permissions, so you can approve reading and logs without publishing or deleting.

For the API side and more debugging prompts, read GoHighLevel Workflows API and MCP: Ask Claude What Broke.

Five safety rules

1. Start in a demo or test sub-account. Move to a live account only when the answers check out.
2. Approve the fewest permissions you need. Our advice is view-only first, edit later.
3. Keep Allow once on while you test. Only set Always allow for read tools you trust.
4. Check results in HighLevel. Treat Claude's answers as a draft, not the record.
5. Only use links from HighLevel's official docs, and never share a Private Integration Token in a chat.

Review or revoke access

You can disconnect at any time in Claude under Customize > Connectors, or review and revoke the connection from your LeadConnector account. Anthropic says you can revoke access by disconnecting the connector in Claude's settings or in the service's own security settings.