WebDecoy's MCP Server: Bot Protection in Your AI Assistant

WebDecoy now has an MCP server. Connect it to Claude, ChatGPT or Codex and your assistant can install WebDecoy in the app you are working on, prove the install is reporting, and answer questions about your sites: is protection actually on, which bots came by, and was that Googlebot really Google.

The server lives at https://mcp.webdecoy.com/mcp. You sign in with your WebDecoy account, choose which sites the assistant may see, and start asking. No API key, no copy and paste.

What you can ask

“Install WebDecoy in this app and check it’s working.” The assistant looks at your project, asks WebDecoy for the install guide that matches your stack (Next.js, Express, Fastify, Hono, plain Node, PHP, WordPress, or the browser script tag), and applies it with your site’s public IDs already filled in. Then it sends WebDecoy’s test request and asks WebDecoy whether the install is reporting. In Claude Code, the WebDecoy plugin’s install skill won’t call the job done until WebDecoy confirms it.

“Is my site protected right now?” WebDecoy answers with the same check the dashboard uses: whether an edge is actually applying your policy, and if not, why. If enforcement is switched on but scoped to no paths, you hear exactly that, not a reassuring green tick.

“What bots hit my site this week, worst first?” Detections come back with what was requested, the threat score, and whether a crawler was who it claimed to be. A request calling itself Bingbot from an address Microsoft does not own is reported as forged.

“Tell me about the actor behind that detection.” The assistant pulls the actor’s history: how long it has been around, how it identified itself, and its most recent requests to your site.

“Add my site and set up decoys.” If you own your organization and allowed setup, the assistant can add a new site, create decoys for it (a hidden link, or a fake API route), and paste the code that hides them where only bots will look. Sites and decoys count toward your plan, and a decoy only records who fetches it.

“What does my policy block?” You get the configured policy for a site, labelled for what it is: configuration, not proof. The protection check is what tells you whether anything enforces it.

Set it up

Pick your assistant. Each one signs you in through your browser the first time you use a WebDecoy tool.

Claude Code

Install the WebDecoy plugin:

/plugin marketplace add WebDecoy/claude-plugins
/plugin install webdecoy@webdecoy

Restart Claude Code, run /mcp and sign in to webdecoy. Then ask it to “install WebDecoy in this app”, or run /webdecoy:install. The plugin adds the server, an install skill that follows WebDecoy’s own guide for your stack, and the install command.

Claude.ai and Claude Desktop

In Settings, Connectors, choose Add custom connector and enter:

https://mcp.webdecoy.com/mcp

ChatGPT

In Settings, Apps and Connectors, turn on developer mode and choose Create. Enter https://mcp.webdecoy.com/mcp as the server URL with OAuth authentication, and under the advanced OAuth settings choose dynamic client registration.

Codex

codex mcp add webdecoy --url https://mcp.webdecoy.com/mcp
codex mcp login webdecoy --scopes mcp:read,mcp:setup,offline_access

Everything else: the WebDecoy CLI

For assistants that run MCP servers as local commands, install the CLI and sign in once:

brew install webdecoy/tap/webdecoy
webdecoy login

webdecoy login shows a code to approve in your browser, and keeps the login in your macOS Keychain or Linux Secret Service, never in a file. Then add the command webdecoy mcp to your assistant. For Claude Desktop that is:

{"mcpServers": {"webdecoy": {"command": "webdecoy", "args": ["mcp"]}}}

The CLI is open source: github.com/WebDecoy/cli.

Then approve your sites

The first time your assistant calls a WebDecoy tool, it gets back a link to approve it. Open the link, tick the sites that assistant may read, and approve. If you want it to add sites, create decoys and script tags, and check installs too, also tick Also allow setup. A site the assistant adds is added to its own access; sites it did not create still need your approval. Every assistant gets its own approval, so the one in your editor can see your staging site without the one in your browser seeing anything at all.

Step by step for each client: Connect an AI assistant.

What the assistant can and cannot do

Handing a security product to an AI assistant only works if the limits are clear, so here they are.

A few more things we built in:

• Captured bot traffic is treated as data. User agents and paths come from bots, and bots write whatever they like, including text aimed at your assistant. Every answer that carries them is marked untrusted, telling the assistant to report that text, never to follow it.
• Revocation is immediate. Disconnect an assistant in Settings, Connected apps and its next request is refused, even if its login has not expired. Remove a teammate from your organization and the same happens to their assistants.
• One assistant cannot crowd out your dashboard. Assistant traffic has its own per-user budget and its own database connections, so a runaway loop in someone’s editor slows down only their own assistants.

Available now, on every plan

The MCP server is live for every WebDecoy account, including Free. If you do not have an account yet, start for free, connect your assistant, and ask it to install WebDecoy for you.