What did HighLevel announce on October 28, 2026?
On October 28, 2026 HighLevel added workflows to its LeadConnector MCP server in private beta. Claude, ChatGPT, Cursor and other AI apps can now find, explain, build, update, publish and delete HighLevel workflows, and read their execution logs and statistics.
How the news broke
The announcement is HighLevel's changelog entry "Workflows Now in MCP: Ask Your AI Assistant to Build, Debug and Publish Workflows", dated October 28, 2026. It opens: "You can now work with your workflows from your AI assistant." HighLevel's support article "Workflows in the LeadConnector MCP Server" was modified the same day, at 2:53 PM. We read both on October 30, 2026.
Workflows are where most of a HighLevel account's automation lives, so this lets an AI app change how an account behaves as well as the records in it. If you are deciding whether to connect an AI app to HighLevel at all, our MCP server development service page explains when the official server is enough and when a custom one is needed.
What can an AI assistant do with HighLevel workflows now?
Seven things: find workflows, read their full setup, build a new one from a description, update and publish, delete, read execution logs, and count enrollments, completions and failures per step. New workflows are always saved as drafts first.
| Action | What HighLevel says it does | What to watch |
|---|---|---|
| Find | Search by name, status, folder or date | Read-only |
| Read | Explains triggers, steps, branches and settings | Read-only |
| Build | Creates a workflow from a description, checks it for errors and saves it as a draft | App steps only work with app accounts already connected in the sub-account |
| Update and publish | Changes a workflow, publishes it, or sets it back to draft | An update replaces the whole workflow; editing a published one turns it to draft |
| Delete | Deletes after an extra confirmation | "Deleting a workflow is permanent and stops everyone currently in it." |
| Execution logs | Who entered, what happened at each step, which steps failed and the error | Up to 30 days per request, 60 days back; live runs only; logs include contact names and emails |
| Statistics | Enrollments, completions and failures per step, by hour, day or week | Read-only |
The server still exposes four tools, list_locations, search_operations, describe_operation and execute_operation, over more than 550 operations. The workflow actions are new operations behind those same four tools. Adding or removing a contact from a workflow is handled through the contact operations, not these workflow actions.
Who can use it, and what does WORKFLOW_API_NOT_ENABLED_FOR_LOCATION mean?
Only sub-accounts that HighLevel has enabled for the Workflows API. Everyone else gets the error WORKFLOW_API_NOT_ENABLED_FOR_LOCATION and has to contact HighLevel support to join the private beta.
Both HighLevel pages describe it as a private beta, enabled for selected sub-accounts. If you see that error, nothing is wrong with your connection: the sub-account simply is not in the beta yet. HighLevel asks that a failed request report include the full error and the traceId if one is shown.
Each request works in one sub-account. Agency connections that span several sub-accounts are, in HighLevel's words, still "rolling out", which is the same limit the main MCP server has had since launch.
How do you connect HighLevel workflows to Claude, ChatGPT or Cursor?
Add the LeadConnector MCP server to your AI app with the address for that app, sign in, pick the sub-account, and approve the workflow permissions. Then start a new chat. There is no separate workflow server.
The changelog lists the addresses: https://services.leadconnectorhq.com/mcp/anthropic/v2 for Claude, https://services.leadconnectorhq.com/mcp/openai/v2/ for ChatGPT (with Developer mode on) and https://services.leadconnectorhq.com/mcp/cursor/v2 for Cursor. The support article says reading, changing, publishing, deleting and reading logs each have their own permission, approved when you connect. It does not name the scope identifiers.
If an existing connection cannot see any workflow actions, reconnect and check that the workflow permissions were approved. Our step-by-step guide to connecting the HighLevel MCP server to Claude and ChatGPT covers sign-in and the common errors.
What are the safety rules and limits?
New workflows are drafts until a separate publish request. Editing a live workflow turns it off until republished. Deletion is permanent. Logs reach back 60 days and skip test runs. Each request covers one sub-account.
A safe path for AI-built workflows
The rule with the biggest business impact is this one from the support article: editing a published workflow sets it to draft, "which turns off its triggers and pauses contacts at their next step". So if an AI app edits your live lead follow-up workflow during the day, new leads stop entering it, and contacts already in it stop moving, until someone publishes it again. An update also replaces the whole workflow, and is rejected if someone is editing it in the builder at the same time.
What this means for your business
Debugging gets faster: anyone on the team can ask an AI app why a workflow failed for a contact and get an answer from the real logs. AI-built drafts help, but a careless edit can pause a live workflow, so keep publishing in human hands.
For most HighLevel users the most valuable part is the read side. Questions that used to mean opening each workflow's history, such as "which step failed most this week, and for whom?", become one question in Claude or ChatGPT. The statistics action answers "how many people finished the onboarding workflow last month?" without building a report.
The write side needs rules. An AI app that can publish and delete workflows has a lot of power over how your account treats leads and customers. Approve only the workflow permissions each person needs, start in a test sub-account, and remember that execution logs include contact names and emails, which then pass through your AI app. If your HighLevel setup handles patient data, check our HIPAA automation guide before connecting any AI app to it.
When you need more control than the official server gives, such as one tool that checks a workflow's failures and also looks up the job in another system, or a rule that blocks any edit to live workflows, that is where a custom server fits. Our GoHighLevel automation team builds the workflows themselves, and our custom MCP server builds add the AI layer on top. The security side is covered in the security risks of MCP servers.
More questions about HighLevel workflows in MCP
Can ChatGPT or Claude build a GoHighLevel workflow for me?
Yes, in sub-accounts that are in HighLevel's private beta. You describe the workflow, the AI app builds it through the LeadConnector MCP server, HighLevel checks it for errors, and it is saved as a draft. Publishing is a separate request. App steps only work with app accounts already connected in that sub-account, so connect those first and review every step before publishing.
Can the AI turn off or break a live workflow?
It can pause one. HighLevel's support article says editing a published workflow sets it to draft, which turns off its triggers and pauses contacts at their next step until it is published again. Deleting is permanent and stops everyone in it, though HighLevel asks for an extra confirmation first. Approve only the permissions each user needs.
Can the AI see why a workflow failed for a contact?
Yes. The execution logs action shows who entered a workflow, what happened at each step, and which steps failed with the error. Each request covers up to 30 days, and logs go back 60 days. Test runs from the workflow builder are not included, so test with a real enrollment if you want the AI to see it.
Does this cost extra in HighLevel?
Neither the changelog entry nor the support article mentions a fee for workflows in the MCP server. You still need an AI app that supports custom MCP connections, such as a Claude or ChatGPT plan, and the sub-account must be in the private beta. Check your own HighLevel billing page, since pricing can change after a beta.
Does this work for agencies across all sub-accounts?
Not yet. Each request works in one sub-account, and HighLevel says agency connections spanning several sub-accounts are rolling out. Until then, an agency connects each sub-account separately, and each one must be enabled for the Workflows API. For agencies with many sub-accounts, a custom server can loop through them with your own rules.
Sources and further reading
- HighLevel changelog, Workflows Now in MCP (October 28, 2026), checked October 30, 2026.
- HighLevel support, Workflows in the LeadConnector MCP Server (modified October 28, 2026), checked October 30, 2026.
- HighLevel, LeadConnector MCP Server developer docs, checked October 30, 2026.
- HighLevel support, How to Setup and Use the HighLevel MCP Server (modified October 29, 2026), checked October 30, 2026.
