One MCP tool that checks supplied text for common LLM writing clichés using the client-side pattern engine from Simon Willison's LLM cliché highlighter. It does not fetch URLs or call any backend service.
highlight-llm-cliches
- Input:
text(plain text, up to 250,000 characters) - Output: one warning per match, with its phrase and sentence context. Each warning also includes the rule, explanation, character offsets, and chain item count where relevant.
The warnings are heuristic writing feedback, not proof that text was AI-generated.
Connect an MCP client to:
https://llm-cliche-highlighter.run.mcp-use.com/mcp
First, run the development server:
npm install
npm run devOpen http://localhost:3000/mcp/inspector with your browser to test your server.
Run npm run typecheck to refresh MCP view types and check the project with its local TypeScript compiler.
Run the automated checks with:
npm test
npm run typecheck
npm run build