Open-source browserd MCP server: 39% faster, 68% cheaper than Playwright MCP in real tests
Delicious_Break5937 · reddit · 2026-11-19
A developer released browserd, a free open-source (MIT) browser MCP server for macOS/Windows, benchmarked against Playwright MCP, chrome-devtools-mcp, and Claude in Chrome.
Design: agents send a list of steps (click, type, screenshot) in one queue call; browserd runs them in order and stops at the first failure, cutting turns drastically — 47–87% fewer tool calls across three long real-world jobs (spreadsheet + deck from ten 10-Ks, a 28-park map, a six-city comparison doc), all on claude-sonnet-5-5 via Claude Code, 9 runs per tool.
Results:
- browserd: 8/9 passed, 193 min, $43.81, 1,107 tool calls
- Playwright MCP: 7/9, 315 min, $66.03, 2,089 calls
- chrome-devtools-mcp: 9/9, 285 min, $138.48, 8,320 calls
- Claude in Chrome: 8/9, 278 min, $76.09, 3,256 calls
That's 31–39% faster and 34–68% cheaper. Caveats disclosed: on four public benchmarks (900 runs) it only ties rivals; no significance test on 36 long-job runs; Chrome-minimized startup favors browserd — worst case, its time lead over chrome-devtools-mcp falls to 9%.
Features: 8 tools; each profile is its own Chrome with persistent logins; agents work in background tabs; a dashboard shows every agent's tabs with one-click handoff for sign-ins/captchas. Motivated by job-application automation and lead research on Chinese sites Claude's web search couldn't open. Security note: it doesn't filter page content, so pages could inject instructions — give each task a profile with only the logins it needs.
More from coding & agent
- Designers need a prototype playground: a copyable prompt-based AI design workflow — round · 2026-11-19
- NVIDIA post-trained Nemotron with Factory's agent harness inside its air-gapped network — maddiehfaulkner · 2026-11-19
- LangChain's Deep Agents hits 50M downloads a year after betting on open agent harness — LangChain · 2026-11-19
- Running 15 AI Agents All Day Is Context-Switching Fatigue, IBM Engineer Explains — TejasKumar_ · 2026-11-19
- Dev's AI-native macOS dock with agent-managed modes flops; pivoting to to-do approach — trq212 · 2026-11-19
- Daiku Gives Your AI a Logged-In Browser So Multiple Agents Can Work Your Tabs in Parallel — TejasKumar_ · 2026-11-19