Tested 6 models: WebMCP is optional for agents but consistently cuts steps and speeds them up
rseroter · x · 2026-09-29
Erik Lieben ran a controlled experiment asking whether Angular apps need WebMCP when browser agents can already operate an accessible page. He built one Angular todo app in two builds — one registering six WebMCP tools following Chrome's tool-design guidance, one without — and tested Claude Haiku/Sonnet/Opus in Claude Code plus GPT-6 Luna, GPT-5.6 Terra and GPT-6 Sol in Codex CLI on the same tasks. Result: all runs succeeded either way, but with tools agents used fewer browser commands in all twelve pairs and finished faster in ten. Agents don't need WebMCP, but a small set of well-designed tools consistently reduces steps and time. He also notes a nuance: an action hidden behind a menu only enters the accessibility tree when opened — exactly where named tools add value.
More from coding & agent
- Scale's Muse agent autonomously files a Verizon complaint from one photo — rickasaurus · 2026-09-29
- Berkeley PhD Candidate Compares Claude Code vs Codex vs Pi: Harness Choice Hits Cost More Than Accuracy — arena · 2026-09-29
- Grok Bot ships voice mode, 1Password and Office integrations in 10 days — pswider · 2026-09-29
- Footnote: Two Cognition FrontierCode Cases Where Subagent Review Caused Timeouts or Scope Creep — cedric_chee · 2026-09-29
- Columbia team uses agents end-to-end for preprint on p53-hormone receptor genomic grammar — anshulkundaje · 2026-09-29
- MCP server for AI-driven project management launches with 8 tools for Claude Desktop — Potential-Art7696 · 2026-09-29