It would be easy to dramatize this as "AI agents can finally tinker with Chrome on their own." The reality is more grounded. Still, Google shipping Chrome DevTools for agents 1.0 as a stable release on May 22, 2026, and bundling it into Antigravity 2.0, is quietly important for indie developers running agent-based workflows.
I have been an indie developer since 2014 (around 50 million cumulative downloads across iOS and Android wallpaper and ambient apps) while running four Lab sites on Antigravity 2.0 and Cloudflare Workers. I tried it on launch day. Here is a concise summary of what changed, how it sits next to similar tools, and the small things worth knowing before you turn it on.
What the stable release actually unlocks
Looking at the official blog post and release notes side by side, the operations now exposed to AI agents through Chrome DevTools for agents 1.0 are:
- User-action replay (click, scroll, form fill, etc.)
- Lighthouse-based site quality audits
- Device, network, and geolocation emulation
- Chrome extension debugging
- Memory leak detection and Performance profiling
- Network request and console message capture
In other words, most of what you used to do by hand inside DevTools is now callable as an API by an AI agent. The Lighthouse integration was flaky during preview, so I never used it in production. With the stable release it is finally consistent enough to wire into CI.
What changes when it lives inside Antigravity 2.0
Until now, driving Chrome from Antigravity 2.0 meant installing an external MCP server separately. With 1.0 bundled into Antigravity 2.0, you get:
- Zero-install access to
chrome-devtoolstools straight from Agent Manager - Per-workspace permission scopes (Project A: Lighthouse only; Project B: also form input)
- DevTools action history integrated into Antigravity's agent logs (easier post-mortem)
For "an indie developer juggling several projects," per-workspace scoping really helps. With four Lab site repositories open at once, restricting permissions to "pre-deploy checklist" workspaces means even a misbehaving agent has limited blast radius.
How it sits next to Playwright MCP and Claude in Chrome
The Hatena Bookmark thread on the release explicitly called out how crowded the space has gotten. Here is the rule of thumb I have converged on.
| Tool | Strength | Typical use |
|---|---|---|
| Chrome DevTools for agents 1.0 | DevTools-equivalent analysis, Lighthouse, memory | Debugging and performance work |
| Playwright MCP | Polished E2E testing and scripting | Test generation, CI integration |
| Claude in Chrome | General-purpose browsing on any page | Research, content, scraping |
| Vertical MCPs (AdMob, etc.) | Industry-specific surface area | Specialist console operations |
Roughly: "want to measure quality / fix a broken page" → Chrome DevTools for agents, "want to write tests" → Playwright, "want an agent to browse like a human" → Claude in Chrome. Trying to consolidate into one ends with every use case being half-served.
Three setup gotchas worth knowing on day one
After spending the launch day on it, three things stood out as easy to trip over.
1. Lock down per-workspace permission profiles
Antigravity 2.0's default permissions sometimes include form submission out of the box. The launch templates are on the loose side, so before going to production, create a "Read-only" profile under Agent Manager → Permissions. I run Read-only on the production workspaces for the four Lab sites.
2. CAPTCHA and bot protection still win
Cloudflare Turnstile and reCAPTCHA pages will simply refuse to load. This is by design. To inspect your own production site you either need to relax Bot Fight Mode temporarily, or allowlist the agent's runtime IPs. I keep my Cloudflare dashboard loose for the four Lab sites only during development hours.
3. Memory analysis takes real time
Continuous Performance monitor for memory leak detection takes 30–60 seconds per pass. Running them in parallel in CI burns budget quickly, so I batch a single nightly job. For an app monetized by AdMob, where memory accumulation directly hurts ad revenue, even monthly is usually enough.
The first command to try
If you want to feel the value in five minutes, here is the minimal first command.
# In Agent Manager inside Antigravity 2.0 (natural language)
"Fetch the Lighthouse scores for this URL, then return only Performance and SEO
in detail. List three concrete improvements: https://example.com"Scores and suggestions come back as JSON and the agent analyzes them on the spot. If you run AdMob, a Performance regression hits revenue directly, and automating this as a nightly job alone changes the experience.
Going deeper
This page covers only the launch news and first-day setup. For a deeper operational design — Lighthouse audits, extension debugging, memory analysis, and headless automation as a single workflow — see Antigravity 2.0 with Chrome DevTools for agents 1.0: a practical workflow design here on Antigravity Lab.
If you call Chrome DevTools for agents from a Claude Code environment, there is a companion piece over on Claude Lab. Reading both should give you the full picture across editors.