WrongBot — a Model Context Protocol server that is confidently, deliberately wrong

What is this?

WrongBot is a small satirical project: a chat bot — and a real Model Context Protocol (MCP) server — that answers every question confidently and incorrectly. It's a joke about AI overconfidence, and a working demo of how easy it is to stand up a real MCP server on Cloudflare Workers.

The chat above runs entirely in your browser, with no server round-trip. The MCP server is the same "wrongness engine," exposed as a tool any MCP-compatible AI client can call directly.

The disclaimer matters: the bot's output — every answer, confidence score, and source — is always fabricated. The text on the rest of this page, describing what WrongBot is and how to use it, is accurate.

Connect an AI client to WrongBot via MCP

Server URL:

Claude Desktop / Claude Code (claude_desktop_config.json):

{
  "mcpServers": {
    "wrongbot": {
      "command": "npx",
      "args": ["mcp-remote", "https://wrongbot.jasperkooij.com/mcp"]
    }
  }
}

Cursor (.cursor/mcp.json):

{
  "mcpServers": {
    "wrongbot": { "url": "https://wrongbot.jasperkooij.com/mcp" }
  }
}

Or try it instantly with no config in the Cloudflare AI Playground — just paste the server URL in.

Once connected, the tool is get_wrong_answer — call it with any question and it returns a confident, fabricated answer, a suspiciously high confidence score, and a fake source. Older clients that haven't moved to Streamable HTTP can use the legacy SSE transport at /sse.

Documentation

The tool: get_wrong_answer

Architecture

wrongbot.jasperkooij.com
├── /                        landing page + docs + demo widget (Workers Assets)
├── /mcp                     MCP server (Streamable HTTP)
├── /sse, /sse/message       MCP server (legacy SSE transport)
└── /api/wrong-answers.json  the raw wrong-answer bank, as JSON

Built on Cloudflare Workers + Durable Objects (required by the MCP Agent SDK for per-session state), the agents package's McpAgent class, @modelcontextprotocol/sdk, and Zod for input validation. No framework, no build step for the frontend.

Adding more wrongness

All the "logic" lives in src/wrong-answers.ts in the repository — a plain object of keyword → wrong-answer bank, plus a list of fake sources. The MCP tool picks up new entries automatically; the browser widget above keeps its own inline copy by design, so it stays a single dependency-free block — keep the two in sync when you edit one.

Machine-readable resources for agents

• /llms.txt — short index for AI agents
• /llms-full.txt — this documentation, as Markdown
• /index.md — this page, as Markdown
• /openapi.json — OpenAPI spec for the REST endpoint below
• /api/wrong-answers.json — the complete wrong-answer bank as JSON
• /.well-known/api-catalog — RFC 9727 API linkset
• /.well-known/ard.json — Agentic Resource Discovery catalog
• /.well-known/mcp/server-card.json — MCP server card
• /sitemap.xml and /robots.txt — with explicit allow rules for named AI crawlers
• /.well-known/security.txt — vulnerability disclosure contact

FAQ

Source

Code, local development, and full deployment instructions: github.com/jasperkooij/wrongbot-mcp (MIT licensed).