Turbo-Browser-MCP is a Playwright-based Model Context Protocol server designed to make browser automation by language-model agents use fewer tool calls, tokens, and inference turns. It combines common browser operations into 10 batched tools, returning compact DOM snapshots and, when requested, only changes since the previous inspection. Its form engine can fill multiple controls in one call, including React-managed inputs, custom widgets, iframes, and open or hooked Shadow DOM. It also supports direct file uploads from disk, parallel tabs with isolated sessions, clean text extraction, screenshots, JavaScript evaluation, network blocking, storage management, and error-focused diagnostics. A default submit guard pauses before consequential form submission for human review.
The project’s local Chromium benchmarks compare these workflows with more granular browser automation. For a 10-field form, it reports a reduction from 11 turns to one, with browser execution dropping from 574 to 82 milliseconds and cumulative token use falling by 98.9%. A dense interface snapshot uses 1,135 tokens rather than 18,654 for HTML; a post-click delta uses 56 tokens rather than 3,582 for a full accessibility rescan. Uploading a 48 KiB PDF by disk path avoids Base64 and cuts the reported input from 44,728 tokens to 85. Other tests show six tabs opening 2.38 times faster when batched, and media blocking eliminating 304 KiB of requests. These measurements frame the central claim: reducing round trips and repeated context can make agent-driven browser tasks faster and cheaper.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.