HighLevel MCP for Anthropic Is Live: What It Actually Does

HighLevel co-founder Shaun Clark announced a dedicated Model Context Protocol (MCP) endpoint built specifically for Anthropic's Claude. It is a new OAuth-based connection that exposes 625 operations across 40 domains of your sub-account to Claude.ai, Claude Code, and Claude Cowork. Here is exactly what it does, how its 5-tool architecture keeps token cost low, and the one question every automation builder is asking: can Claude actually build your workflows?

What is the HighLevel MCP for Anthropic?

The Model Context Protocol (MCP) is an open standard that lets an AI model discover and call external tools through one consistent interface, instead of every integration inventing its own. This release is HighLevel's dedicated MCP server built for Anthropic's Claude. Previously, AI models using the universal endpoint only saw a narrow, limited-scope surface. The Anthropic endpoint uses a dedicated OAuth application that exposes the widest scope set on the platform, spanning read, write, and delete actions across the CRM.

The endpoint is live at services.leadconnectorhq.com/mcp/anthropic/v2. That URL is the first of a new per-client pattern, /mcp/{client}/v2: HighLevel deploys a unique path for each AI architecture, so the scopes you grant Claude stay isolated from any other AI tool you connect later. It is fully compatible with Claude.ai, Claude Code, and Claude Cowork.

The 5-tool architecture: small context, massive power

Exposing hundreds of API endpoints directly to a language model usually floods its context window, which raises token cost and drops execution accuracy. HighLevel solved that with a hyper-efficient 5-tool facade. Instead of loading 625 operation schemas up front, Claude searches an internal operations registry, pulls the exact documentation for the call it needs, and executes it dynamically.

1. search - find records and content across the sub-account.
2. fetch - pull a specific record or resource by reference.
3. search_operations - search the hosted registry of available API operations.
4. describe_operation - retrieve the exact schema and parameters for one operation.
5. execute_operation - run the operation with the arguments Claude assembled.

625 operations across 40 domains

The dedicated Anthropic OAuth app exposes the widest scope set available on the platform. The split below is what the new endpoint unlocks compared to the focused, limited-scope surface older universal connections were stuck with.

Does the HighLevel MCP create workflows?

This is the question worth slowing down on, because 244 write operations sounds like Claude can now build your automations for you. It cannot, and the reason is structural. HighLevel states that every executed action pulls from its latest v3 API engines, and the v3 public API does not expose workflow authoring.

On the v3 public surface, the Workflows API does exactly two things: it lists the workflows in a sub-account, and it adds or removes contacts from an existing workflow. That means Claude can read which workflows you have and enroll a contact into one that already exists, but it cannot create a workflow, add a trigger, drop in actions, build if/else branches, or edit the logic inside one. Workflow authoring simply is not part of the operation registry the MCP is built on.

That is not a knock on the release. Creating contacts and opportunities, moving pipeline stages, booking and rescheduling appointments, sending messages, managing tags and custom fields, and enrolling people into existing workflows is a large amount of real operational work, now drivable in plain language. It just is not the same thing as designing the automation logic.

How secure is the HighLevel MCP?

• Granular, user-controlled scopes: you choose exactly which permissions to grant on the standard OAuth consent screen when you connect Claude.
• Sub-account boundaries: a connection is scoped strictly to one sub-account (location), never agency-wide, so client data stays isolated.
• Modern auth stack: Authorization Code with PKCE and rolling refresh tokens managed through a secrets manager.
• RBAC enforcement: every executed action respects Role-Based Access Control. If your active user session does not have permission to perform an action in the web app, Claude is blocked from performing it too.

Does the old HighLevel MCP endpoint still work?

Yes. The original universal /mcp/ endpoint remains fully active and supported for any standard MCP client using a Private Integration Token (PIT), but it continues to run on the legacy, limited-scope surface. The new per-client OAuth framework is where the full scope set lives. HighLevel says it is actively expanding that framework, with dedicated endpoints for OpenAI, Cursor, Windsurf, and VS Code planned for later this month.

How to connect Claude to HighLevel

1. Open your Claude client (Claude.ai, Claude Code, or Claude Cowork) and add a new MCP connector.
2. Point it at the endpoint: services.leadconnectorhq.com/mcp/anthropic/v2.
3. Complete the OAuth consent flow and pick the exact scopes you want Claude to have.
4. Confirm the connection is scoped to the single sub-account you intend, not agency-wide.
5. Test with a read action first (list contacts) before granting write and delete usage in a live account.

What this means for your agency

The Anthropic MCP turns Claude into a capable operator over your HighLevel data: reading, creating, updating, and cleaning records across 40 domains, safely gated by the same RBAC that governs your team. What it is not, yet, is an automation builder. Knowing that line - operate data with the MCP, build workflows in the builder - keeps your expectations accurate before you wire it into a client account. Need help connecting Claude to HighLevel safely, or building the workflows the MCP will run against? That is exactly what we do at BrewedOps.