ask-qwksearch-mcp-server

Create a file in src/tools exporting a register function that calls server.registerTool with a description, zod inputSchema, and handler, then import it in src/index.ts. Return isError content instead of throwing, since a thrown error kills the stdio session.

Add an entry to claude_desktop_config.json under mcpServers with command bunx and args qwksearch-mcp, plus an env block for SCRAPER_URL and SCRAPER_API_KEY. The client must use bun, not node, because the entry point is TypeScript.

The tool refuses unauthenticated calls when SCRAPER_API_KEY is not set. Define it in the MCP client's env block rather than only in your shell, and ensure it matches the worker's SCRAPER_API_KEY binding to avoid 401 responses.

The client is running node against a TypeScript entry file. The server has no build step or dist output, so it must be executed with bun or bunx, which run TypeScript directly.

The category parameter maps to engine groups, and some enum values like science or files have no engines registered under them. Retry with general or academic, which have registered engines.

No, all three tools return a single pre-formatted text block by design. If you need structured objects, import the underlying search-web-api or extract-webpage packages directly instead of going through the MCP layer.