Read blocked web articles through a browser-control MCP

The poster used Claude with a browser-control MCP that can navigate, click, type and run JavaScript to search and read articles it otherwise could not access. · Nov 7, 2026

Done withCLClaude

What they did

The poster had built MCP servers for Claude, including one that could control a browser by navigating, scrolling, clicking, filling form fields, and triggering page JavaScript. They used it as a fallback for web research in a job-search workflow because normal Claude web fetches were often blocked; when one site showed a human check, Claude refused to keep using that browser MCP by default and the person had to consider handing steps off to another model instead.

How it went

For about a year, the browser-control MCP had let Claude search and read articles that its normal web tools could not access. The failure was that, after one human-check incident, Claude would not adopt that MCP as the standard approach for similar future tasks.

Worth knowing

The key catch is reliability: the setup had worked for roughly a year, but one guardrail-triggering site could poison the session and force the poster toward a more cumbersome multi-model orchestrator.

Source: Reddit · Nov 7, 2026