GoHighLevel API文档MCP服务器

ghl-api-mcp

ghl-api-mcp is an MCP stdio server for GoHighLevel documentation. It lets agents search docs, list endpoints, and fetch focused API details without loading large markdown files into prompt context.

This repo does not include a runnable GoHighLevel OAuth/SSO sample app; it contains documentation and the MCP docs server that indexes that documentation.

What It Does

The server indexes markdown files under docs/ and exposes MCP tools for:

• list_docs
• search_docs
• list_endpoints
• get_endpoint_details
• get_section
• get_document
• reload_docs

The current corpus includes API review docs, implementation guides, and operational notes such as OAuth/SSO integration details and rate-limit guidance. To see exactly what is loaded, run list_docs.

list_docs labels each file as either a reference or guide. Use get_document when an agent needs the full body of a longer procedural document instead of a single extracted section.

For search_docs, list_endpoints, get_endpoint_details, and get_section, you can pass an optional whitelabel_domain like api.example.com or https://api.example.com. Returned endpoint URLs and embedded example URLs will be rewritten to that domain.

Requirements

• Node.js >=20 for local usage
• Docker, if you want to run the containerized version

Quick Start

Local

npm install
npm start

Or run the CLI entrypoint directly:

npm install
./bin/ghl-api-mcp.js

Important: this is a stdio MCP server. When you run it successfully, it stays attached and waits for client requests. It is not an HTTP server and does not open a port.

Docker

docker build -t ghl-api-mcp .
docker run --rm -i ghl-api-mcp

MCP Client Configuration

Local Repo Checkout

Use this when the repository already exists on disk:

{
  "mcpServers": {
    "ghl-docs": {
      "command": "node",
      "args": ["/absolute/path/to/ghl-api-mcp/src/server.js"]
    }
  }
}

This is the preferred setup for local agents. It reads the repo's docs/ directory directly, so docs-only edits do not require rebuilding Docker images and MCP clients do not create Docker containers.

Local Repo Checkout With External Docs Directory

{
  "mcpServers": {
    "ghl-docs": {
      "command": "node",
      "args": ["/absolute/path/to/ghl-api-mcp/src/server.js"],
      "env": {
        "GHL_DOCS_DIR": "/absolute/path/to/docs"
      }
    }
  }
}

Docker

Use Docker only when the repo is not available to the MCP client. For a local checkout, prefer the Node configuration above so client restarts cannot leave old MCP containers running.

{
  "mcpServers": {
    "ghl-docs": {
      "command": "docker",
      "args": ["run", "--rm", "-i", "ghl-api-mcp"]
    }
  }
}

Docker With External Docs Directory

{
  "mcpServers": {
    "ghl-docs": {
      "command": "docker",
      "args": [
        "run",
        "--rm",
        "-i",
        "-v",
        "/absolute/path/to/docs:/docs:ro",
        "-e",
        "GHL_DOCS_DIR=/docs",
        "ghl-api-mcp"
      ]
    }
  }
}

Document Loading

• Any .md file under docs/ is indexed automatically
• If the server is already running, call reload_docs after changing docs on disk
• If you add docs to the repo and run via Docker without a mounted docs directory, rebuild the image so the container includes the new files

Development

npm install
npm test

Scraping Workflow

Automatic full-corpus synchronization

The documentation corpus is synchronized from two official sources:

1. The rendered HighLevel API documentation website is the freshness authority. Its sitemap currently exposes the canonical, unversioned pages, which are converted to Markdown under docs/website/.
2. HighLevel's GoHighLevel/highlevel-api-docs repository supplies structured OpenAPI schemas and source Markdown under docs/generated/.

This split is intentional: the website can be newer than the public source repository, while the repository often contains richer request and response schemas than the rendered API pages. Knowledge-base document and section lookups prefer website snapshots; exact endpoint lookups prefer the structured v3/v2 OpenAPI documents.

Run a synchronization locally:

npm run docs:sync

Check whether the committed corpus is current without changing files:

npm run docs:check

Hand-authored review documents in docs/ are never overwritten. Both syncs are deterministic and record source/output hashes in docs/website/.ghl-website-manifest.json and docs/generated/.ghl-sync-manifest.json. Normal local website syncs cache pages whose non-empty sitemap timestamp has not changed. Pages without a sitemap timestamp are fetched on every sync so new content is not cached indefinitely. This includes SDK and Marketplace CLI guides as well as API reference pages. Deleted sitemap routes are removed automatically.

npm run docs:check forces a complete website revalidation, even when sitemap timestamps are unchanged. The scheduled workflow does the same, preventing an incorrect or missing lastmod value from hiding website drift.

The Sync HighLevel API documentation GitHub Actions workflow runs every Monday and can also be started manually. When either official source changes, it runs the test suite and opens or updates a pull request from automation/sync-highlevel-api-docs. Repository Actions settings must allow GitHub Actions to create pull requests.

You can pin a different upstream branch or tag with GHL_DOCS_REF, or test a local checkout without network access:

node scripts/sync-ghl-docs.js --source-dir /path/to/highlevel-api-docs

To refresh only the rendered website snapshots, or inspect one page during development:

npm run docs:sync:website
node scripts/sync-ghl-website.js \
  --url https://marketplace.gohighlevel.com/docs/other/user-context-marketplace-apps \
  --output /tmp/ghl-website-page

Single-page browser scraping

Use the browser-backed scraper when a GHL doc links to richer ClickUp content:

npm run scrape:page -- \
  --url https://marketplace.gohighlevel.com/docs/ghl/contacts/search-contacts-advanced \
  --output docs/generated/search-contacts-advanced.md

Browser selection order:

1. Chrome
2. Chrome Canary
3. Zen

You can override the browser with --browser, --executablePath, or PUPPETEER_EXECUTABLE_PATH.