Access, auth, and limits
The GoHighLevel MCP server lives at https://services.leadconnectorhq.com/mcp/ and speaks plain HTTP (POST). Auth is a Private Integration Token (PIT) passed as a Bearer token in the Authorization header. The piece I care about most is the scope model: tokens carry granular, permission-based scopes, so I grant an agent exactly the tools it needs (read contacts, write opportunities, send messages) and nothing else. For write access into a live client account, that least-privilege control is the part you do not want to skip.
Setup is short: generate a Private Integration Token in the account settings, pick the scopes, drop the endpoint and Bearer token into your MCP client config, and connect. The server launched with 36 tools across Calendar, Contacts, Conversations and messaging, Location data, Opportunities and pipelines, Payments, Blog, Email templates, and Social posting, with a published roadmap to 250+ tools. No SDK install, no scraping layer, no Zapier middleman: the agent reads and writes GoHighLevel data straight through the protocol.
Where it works and what to build
Because the server is a standard HTTP MCP endpoint, it connects to any client that speaks the HTTP transport. GoHighLevel calls out Cursor, Windsurf, and the OpenAI Playground specifically, and in practice Claude Desktop and Claude Code connect the same way since they support remote HTTP MCP servers. That is the upside of an HTTP server over a stdio-only one: I wire the endpoint and Bearer token once and most modern agent clients can use it.
The workflows worth running once it is connected are the agency-operations ones. Have an agent pull pipeline opportunities and update deal stages in bulk after a sales sync, so the pipeline reflects reality without anyone clicking through cards. Point an agent at incoming conversations to triage and draft replies across SMS, email, and chat from one place, then queue them for human send. Or let an agent create and schedule social posts and blog drafts from a content brief, then push them straight into the right sub-account. Each of these is a task an account manager does by hand today, and each maps directly onto the launched tool set rather than something on the roadmap.
Which AI clients can use GoHighLevel MCP?
Not every AI assistant supports MCP natively. Per-client picture for GoHighLevel:
| Client | Support | Notes |
|---|---|---|
| Cursor | Native MCP | Called out by GoHighLevel as a supported client; connect the HTTP MCP endpoint with a Private Integration Token as Bearer. |
| Windsurf | Native MCP | Called out by GoHighLevel as a supported client over the HTTP MCP transport. |
| ChatGPT | Via connectors | GoHighLevel highlights the OpenAI Playground; in the ChatGPT app connect via a custom connector pointed at the HTTP MCP endpoint. |
| Claude (Desktop/Web) | Native MCP | Works as a remote HTTP MCP server: add the endpoint and Bearer token in Claude Desktop connector settings. |
| Claude Code | Native MCP | Add the HTTP MCP endpoint with the Private Integration Token; any HTTP-supporting MCP client connects the same way. |
How to connect GoHighLevel MCP to Claude Desktop
Add this block to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows), then restart Claude Desktop. Cursor and Codex use a similar schema.
{
"mcpServers": {
"gohighlevel": {
"command": "npx",
"args": ["mcp-remote", "<YOUR_MCP_URL>"]
}
}
}
// HTTP-transport MCP. Get your <YOUR_MCP_URL> from the vendor dashboard.
// Server source: https://marketplace.gohighlevel.com/docs/other/mcp/
// Claude Code (one-liner): claude mcp add gohighlevel <YOUR_MCP_URL> --transport httpSome MCP servers require an API key. Check the source page for setup: marketplace.gohighlevel.com/docs/other/mcp.
Does GoHighLevel MCP cost extra?
GoHighLevel ships an official MCP server that uses your existing subscription credentials.
- No separate MCP licence to buy. You pay for your GoHighLevel plan, the MCP server queries that account.
- Whether it's gated to a specific tier wasn't extracted automatically — verify the access requirements on the MCP page or pricing docs.
GoHighLevel subscription pricing
| Plan | Monthly | Annual / mo |
|---|---|---|
| Starter | $97 | — |
| Unlimited | $297 | — |
| Agency Pro | $497 | — |
| Enterprise | Contact sales | — |
Source: gohighlevel.com/pricing. Verified 2026-06-07.
GoHighLevel also publishes a REST API
GoHighLevel publishes an API, but per-tier gating wasn't extracted automatically.
What you can do with GoHighLevel from an AI agent
Three shapes of agent workflow. Pick the row that matches how you actually use Claude / Cursor / Codex day to day.
Skip the dashboard, ask Claude
- "What changed this week?" answered in seconds
- Quick one-off questions without leaving the editor
- Pipe GoHighLevel data into your personal notes / docs without writing a script
Run client analysis at agent speed
- Hand the MCP server to your team — every consultant can ask Claude about any client's GoHighLevel account
- Multi-account workflows without context-switching the GoHighLevel UI
- Reports you'd normally spend hours on, done before standup
Production agent workflows
- Embed in your internal Codex / Cursor agents that triage GoHighLevel data daily
- Combine with other MCP servers in a multi-step agent chain
- Constrain agents to read-only roles to keep production GoHighLevel safe
Limits and gotchas
- Official MCP server — published by GoHighLevel themselves, so it should track product changes.
- Fallback: the REST API works for any flow the MCP server doesn't cover.
- OpenAPI spec is published — easy to validate which endpoints the MCP server actually exposes vs the full API surface.
- Rate limits: MCP calls count against your underlying GoHighLevel API quota. Burst usage from a curious agent can drain a daily allowance fast.
Agent-readiness verdict
Score: 8/10 on MCP-readiness. GoHighLevel ships a first-party MCP server at a single HTTP endpoint with token-scoped, permission-based access and 36 tools live across the parts of the platform agencies actually touch (contacts, conversations, pipelines, calendars, payments). The roadmap to 250+ tools tells me this is a real commitment, not a checkbox. I am holding it at 8 rather than higher because the launch tool count is a slice of the full API surface (so some workflows still need direct REST calls), and the client story leans on generic HTTP MCP support rather than a long list of vendor-tested native integrations.
Strong agent fit with an official MCP server. HighLevel runs a hosted MCP server on its own marketplace domain exposing 36 tools across contacts, conversations, opportunities, payments, and calendars, authenticated with private integration tokens. Paired with published OpenAPI specs, that gives agents a documented, first-party path to automate an account. The main gap is structured-output reliability, so expect some custom error handling.
Scored 2026-06-07 by Joonas (TMB).
GoHighLevel MCP & API FAQ
Is the GoHighLevel MCP server official?+
Can I use the GoHighLevel MCP with Cursor or Codex?+
Does GoHighLevel have a public API?+
Does GoHighLevel publish an OpenAPI / Swagger spec?+
Sources
- GoHighLevel official site: https://www.gohighlevel.com
- MCP server page: https://marketplace.gohighlevel.com/docs/other/mcp/
- API docs: https://marketplace.gohighlevel.com/docs/
- Pricing source: https://www.gohighlevel.com/pricing (verified 2026-06-07)
Data verified by Joonas on the dates shown. MCP server status auto-rechecked weekly.
10+ years in digital marketing. I review marketing software for AI-stack fit: real pricing, MCP and API support, and how cleanly each tool drops into an AI agent workflow, cross-checked against verified data and real user feedback.