Claude Desktop + your local sites: connecting an MCP server that can ship

Claude Desktop connects to a local MCP server the same way it connects to any HTTP MCP endpoint: you point it at a URL, and it discovers the tools that server exposes. Point it at ReDock on localhost:47200 and Claude can do real work — create a local site, start and stop it, author WordPress through Gutenberg blocks, share it over a Cloudflare tunnel, and deploy to Cloudflare Workers. Everything stays on your machine until you decide otherwise.

The difference between a chat that talks about your site and one that can change it is the server behind it. Here’s how to give Claude Desktop hands.

What “connect Claude Desktop to an MCP server” means

Claude Desktop is an MCP client. On its own it can reason and write; it can’t touch your filesystem or run your sites. An MCP server is what supplies capabilities — a set of tools Claude can call.

ReDock is that server, and it’s already running. It’s a native macOS app (macOS 12+, Tauri and Rust, not Electron) that hosts your local sites and serves its entire surface as MCP tools. Because it listens on an HTTP port rather than being spawned per-client, connecting is just naming the endpoint — no command line, no config file to hand-edit. Easy MCP setup covers the zero-config idea in full.

Connecting, step by step

1. Install ReDock and open it. The app and its MCP server start together. See the install guide.
2. Create a site — WordPress, Astro, Next.js, Node, Bun, PHP, whatever you’re building. It comes up at https://name.test with a locally trusted cert. The first-site walkthrough has the details.
3. Add the MCP server in Claude Desktop, pointing at localhost:47200.
4. Confirm the tools appear. Claude lists what ReDock exposes; now it can act.

That’s the whole setup. The full tools list shows what Claude can reach once connected.

What Claude can do once it’s connected

This is where “a server that can ship” earns the phrase. With ReDock’s tools, Claude Desktop can:

• Stand up and run sites. Create a site, start it, stop it — with status read from the real process, not assumed.
• Author WordPress properly. Build pages from real Gutenberg blocks and patterns, and magic-login to wp-admin when a human needs to look. See WordPress through MCP.
• Share. QuickShare for a throwaway public URL, or a named redock.xyz subdomain for a stable one, over Cloudflare tunnels.
• Deploy. One-click Cloudflare Workers deploys, with custom domains. Deploy a website for free covers it.

No API keys are required for the local agent path — ReDock ships a built-in agent and a local-LLM catalog. Claude Desktop simply drives the same tool surface.

Why honest state matters for a desktop agent

When you’re chatting in Claude Desktop, you don’t see the terminal. So the tool results are your only window into what’s real. ReDock maps every status — running, stopped, shared — from the actual process state. If a start fails, the tool says so, and Claude can react instead of narrating a success that didn’t happen. This is the quiet feature that makes an autonomous desktop workflow trustworthy.

Because the server is HTTP, this same session can coexist with others. Your editor’s agent can be connected to the same localhost:47200 at the same time — Cursor + MCP shows that side.

FAQ

Does Claude Desktop need the terminal to use ReDock?

No. ReDock runs sites, certificates, and the MCP server without a terminal. Claude Desktop just connects to the endpoint and calls tools.

Can Claude Desktop actually deploy a site, or only edit it?

Both. The tool surface includes sharing over Cloudflare tunnels and one-click deploys to Cloudflare Workers, so Claude can take a local site all the way to a public URL.

Do I need an API key?

Not for the local path. ReDock includes a built-in agent and a local-LLM catalog, so you can work without wiring external keys.

Give Claude Desktop a local web stack it can actually run and ship — download ReDock.