Build MCP Servers by Asking — GetMCP’s Built-in MCP Server

GetMCP turns your APIs into MCP servers. As of 1.4.0, it applies that idea to itself: GetMCP is an MCP server too. Connect your AI client to your install, and the builder becomes something you talk to. Mid-conversation with Claude about an API? Ask it to build the server. On your phone with ChatGPT? Add a tool to production without opening the dashboard. That’s Manage with AI — creating MCP servers and tools on the go, from whatever client you’re already in.

What your AI client can do

The built-in server exposes the builder’s core as MCP tools: list and inspect your servers and their tools, create and update servers, create, update and delete tools — and, the one you’ll use most, import a tool from a cURL command. Copy a request out of any API’s documentation, paste it into chat, and ask. The AI hands it to GetMCP, and a working, schema’d tool appears on your server.

It handles awkward APIs, too. Your AI can declare where each tool argument travels — path, query, body or header — so an API that, say, deletes rows with a JSON body on a DELETE request works exactly as documented. And for region-specific platforms like Mailchimp or Freshdesk, it sets the data-centre or subdomain up as a connection variable the whole server shares, leaving you one blank to fill.

Connecting takes a minute

Open Manage with AI in your GetMCP dashboard, and it walks you through connecting the client you use — Claude, Cursor, ChatGPT, Windsurf or VS Code — with a readiness check that tells you if anything needs setting up first. In Claude, it’s a custom connector pointed at your install’s endpoint:

https://your-site.com/mcp/getmcp

Authentication is OAuth, and only OAuth — you approve the connection by signing in to your own install in the browser, on a consent screen that names exactly what you’re granting. There is no API key to copy, nothing to paste into a config file, no secret that can leak from a dotfile.

Guardrails, because this is your production config

• Your permissions, re-checked on every call. A token proves who authorized it — what it may do is checked against your account live, so demoting a user revokes their client’s reach immediately.
• No server deletion, by design. The one destructive tool — deleting a single tool — requires explicit confirmation from the AI.
• Safety hints on every tool tell the AI which operations read and which ones write before it runs anything.
• Full visibility. Manage with AI lists every connected client — who authorized it, when, and until when — with a revoke button per client, next to revoke-all.

Try this first

“Create an MCP server called Orders for our internal API, then import this cURL as a tool: curl https://api.northwind.io/v2/orders?status=open -H 'Authorization: Bearer …'. Name the tool List Open Orders.”

Thirty seconds later there’s a server with a working tool in your dashboard, ready to test in the playground. Manage with AI works identically on the WordPress plugin and the standalone app — the full walkthrough, including per-client setup steps, is in the Manage GetMCP with AI guide.