Serve MCP Documentation

What is Serve MCP?

Serve MCP is a hosted Model Context Protocol (MCP) server platform. You connect your business tools — Shopify, GitHub, Klaviyo, a UniFi network, and more — to a Serve MCP workspace. We give you a single hosted MCP endpoint URL and a bearer token; paste those into Claude, Cursor, or any MCP-compatible client, and your AI assistant can immediately call tools against your connected services.

Quick start

1. Create a Serve MCP account and a workspace.
2. From your workspace, go to Connections and add a service (e.g. Shopify, Klaviyo, UniFi). You'll authorize it via OAuth or by entering an API key/credential, depending on the service.
3. Go to MCP Endpoint in your dashboard to get your endpoint URL and generate a bearer token.
4. Paste the URL and token into your AI client's MCP configuration (examples below).

Connecting your AI client

Step-by-step for each client: the setup guide covers Claude on web, desktop, mobile and Claude Code, plus ChatGPT and Cursor. On Claude you paste one URL and sign in — no token needed. The config below is for clients that read a JSON file.

Your endpoint URL is workspace-specific and looks like https://servemcp.com/api/mcp/your-workspace. Every request must include your bearer token in the Authorization header.

A ?token= query parameter also works, but treat it as a last resort for clients that cannot set headers. URLs are written to web server and CDN access logs in plaintext and can leak through the Referer header, so a token sent this way should be considered exposed — revoke and replace it once the client is fixed.

{
  "mcpServers": {
    "serve-mcp": {
      "url": "https://servemcp.com/api/mcp/your-workspace",
      "transport": "http",
      "headers": {
        "Authorization": "Bearer <YOUR_TOKEN>"
      }
    }
  }
}

curl -X POST https://servemcp.com/api/mcp/your-workspace \
  -H "Authorization: Bearer <YOUR_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list","params":{}}'

Both the endpoint URL and a token generator are available on the MCP Endpoint page inside your workspace dashboard — you never need to construct or copy them from anywhere else.

Available integrations

Each integration exposes a curated set of tools — never a generic API passthrough. Tools are individually enabled per connection, and every tool is annotated as read-only or write so your AI client (and you) always know what it can do before calling it.

Every tool name and description, machine-readable: /llms-full.txt.

Setting up a UniFi Network connection

UniFi is a direct-connection, credential-based integration (no OAuth): you provide your UniFi console's public URL and an API key generated from the console. Serve MCP tests the connection, discovers the sites visible to that key, and lets you choose exactly which sites — and how much detail (Basic or Advanced) — your AI assistant can access. You can revise that site selection at any time from the connection's Manage Sites button. All UniFi tools are read-only.

Security

• All connection credentials are encrypted at rest (AES-256-GCM) and are never returned by any API response or MCP tool call.
• MCP tool calls are logged (tool name, status, duration) for auditing — we don't log the arguments you pass or the data returned from your connected services. When a call fails we store the error message the provider returned, which can occasionally quote a value from the request.
• Every tool is annotated read-only or write, and destructive actions (delete, remove, revoke, block) are flagged distinctly from ordinary creates/updates.
• Bearer tokens can be revoked at any time from the MCP Endpoint page, and take effect immediately.
• Integrations with site- or resource-level scoping (like UniFi) enforce that scoping on every tool call server-side, not just in the UI.

For the full details, see our Privacy Policy.

Need help?

Check the FAQ for common questions, or contact us directly if you run into an issue connecting a service or an AI client.