GoHighLevel MCP Server
Enables AI assistants to interact with GoHighLevel CRM through MCP tools, with support for OAuth 2.1 remote deployment and agency multi-sub-account mode.
README
GoHighLevel MCP Server
Model Context Protocol server for GoHighLevel. It exposes GHL API operations as MCP tools over stdio, Streamable HTTP, legacy SSE, and optional MCP Apps.
This fork adds two things on top of upstream:
- OAuth 2.1 for remote deployment — a self-contained authorization server (dynamic client registration, PKCE, refresh tokens, password-gated consent) that secures the Streamable HTTP
/mcpendpoint so it can be added directly to claude.ai as a custom connector. Enable it by settingMCP_OAUTH_PASSWORD.- Agency (Company) multi-sub-account mode — one agency OAuth credential, and you choose which sub-account to work in at runtime via
ghl_list_subaccounts/ghl_select_subaccount/ghl_current_subaccount. Per-location tokens are minted on demand. Enable it with theGHL_AGENCY_*env vars.Deploying it as a remote connector (Railway + Claude): see DEPLOY-RAILWAY.md.
Neither feature changes the default single-sub-account, local/stdio behavior — both are opt-in via environment variables.
New here? Start with QUICKSTART.md.
Using an AI/dev agent? Give it AGENT_SETUP.md and say: "Set this up for my MCP client using the curated profile. Ask me for credentials if needed. Do not run write tools."
5-Minute Quickstart
Requirements:
- Node 20+
- A GoHighLevel private integration token or OAuth access token
- A GoHighLevel Location ID
npm install
cp .env.example .env
npm run build
npm run doctor
npm run configure:codex
Add your credentials to .env:
GHL_API_KEY=your_private_integration_api_key
GHL_LOCATION_ID=your_location_id
GHL_BASE_URL=https://services.leadconnectorhq.com
GHL_API_VERSION=2023-02-21
GHL_API_VERSION=2023-02-21 is the current HighLevel API Version header used by official docs. It is not the project year, and it should not be changed to 2026 unless HighLevel publishes a new required API version.
Then verify live auth:
npm run auth-check
Setup Commands
npm run setup # Create .env if needed, build, and print next steps
npm run first-run # One-command beginner setup/readiness flow
npm run connect # Setup plus client config generation
npm run ready # Fast readiness check
npm run demo # Print MCP Apps demo preview instructions
npm run explain-error -- "Location is not active"
npm run doctor # Human-readable setup check
npm run doctor -- --json # Agent-readable setup check
npm run agent:check # Safe validation for AI/dev agents
npm run auth-check # Read-only GHL token/location check
Missing credentials are reported as needsHumanAction, not as a broken install. This lets agents build and configure the repo without inventing secrets.
MCP Client Config
Beginner configs use GHL_TOOL_PROFILE=curated so agents see the high-level workflow tools first.
npm run configure:codex
npm run configure:claude
npm run configure:cursor
npm run configure:windsurf
Advanced examples:
node scripts/ghl-mcp.mjs configure codex --profile stable
node scripts/ghl-mcp.mjs configure codex --profile full
node scripts/ghl-mcp.mjs configure codex --profile curated --json
Tool Profiles
curated- recommended for agents; high-level CRM workflows with confirmation queues.stable- production-friendly; official, supplemental, curated, and legacy-compatible tools.full- everything.official- official OpenAPI and live-docs supplemental tools.raw- endpoint-level tools only.
Run
npm run start:stdio # Desktop MCP clients
npm run start:http # Streamable HTTP at /mcp
npm run start:legacy # Legacy SSE at /sse
HTTP also exposes:
GET /healthGET /capabilitiesGET /toolsPOST /executePOST /tools/call
MCP Apps
npm run apps:setup
npm run apps:preview
Open http://localhost:3001/preview. Without GHL credentials, the apps use preview/demo states and tell you exactly which env vars are missing.
Tool Discovery
npm run tools:list
npm run tools:list -- --search contacts
npm run tools:list -- --category contacts
npm run tools:list -- --stability official
npm run tools:list -- --access write
npm run tools:list -- --destructive
npm run tools:explorer
The static explorer is docs/tool-explorer.html.
High-Level Agent Tools
Start agents with the curated profile and prefer these high-level tools before raw endpoints:
crm_location_overviewcrm_daily_briefingcrm_search_everythingcrm_next_best_actionscrm_get_next_pagecrm_prepare_contact_followupcrm_prepare_lead_reactivationcrm_prepare_missed_call_responsecrm_prepare_pipeline_cleanupcrm_prepare_review_request_batchcrm_prepare_invoice_followup
Docs
- Update Log
- Setup
- Usage
- Clients
- Tool Profiles
- Recipes
- Safety
- Troubleshooting
- Deployment
- Development
- API Coverage
- Companion Tooling
Update History
| Date | Update # | Included |
|---|---|---|
| 2026-06-11 | 2 | Simplicity and power layer: easy setup commands, safe config writing, grouped live smoke checks, and high-level curated CRM agent tools. See UPDATE_LOG.md for the full permanent update description. |
| 2026-06-11 | 1 | Onboarding and agent setup overhaul. See UPDATE_LOG.md for the full permanent update description. |
API Coverage
- Official GHL endpoints parsed:
590 - Official endpoint coverage:
590 / 590 - Generated official endpoint tools:
238 - MCP tools in registry:
848 - Local-only endpoint references tracked for review:
253
Generated coverage artifacts live in docs/. Run npm run scan:ghl-api only when intentionally refreshing API coverage.
Safety
.envis ignored and must never be committed.test-toolrefuses write/destructive tools unless--confirmis supplied.- Curated workflow tools stage confirmation queues for writes.
- Use
curatedfor beginners andstablefor production.
Credits
This project is a fork of and builds on BusyBee3333/Go-High-Level-MCP-2026-Complete, which provides the underlying GoHighLevel MCP server and tool coverage. That project is licensed under ISC; see LICENSE for the original copyright notice.
This fork adds the OAuth 2.1 remote-connector layer and the agency (Company) multi-sub-account mode described above.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。