Tools Bench.

The official Model Context Protocol server from Google's Chrome DevTools team gives AI coding agents live access to a real Chrome browser for performance traces, DOM inspection, network analysis, and console debugging. It's one of the most durable signals in the agent-tooling space right now: enormous open-source adoption, heavy real-world usage, and first-class support across every major coding agent.

• ~45,600 GitHub stars, maintained directly by Google's Chrome DevTools team
• ~1.4M npm downloads per week; flagged as a key ecosystem project by dependency tooling
• 56 releases in under a year, latest v1.5.0 on July 3, 2026
• Official integrations across Claude Code, Cursor, VS Code/Copilot, Gemini, Cline, Windsurf and more
• Front-page Hacker News feature reached 604 points and 234 comments

• "gives your AI 'eyes' in the browser so it's no longer coding with a blindfold on"Dev.to article

• "a real game-changer for anyone who writes code for the web"Medium article

• "I had success using it to debug frontend performance issues"HN comment

• "the LLM captured traces and isolated bottlenecks and repaints"HN comment

• "really clutters your context - Playwright-cli is more efficient"HN comment

• "MCP times out when the browser has the MetaMask extension"GitHub issue

• "memory leak: grows ~13 MB/min in autoConnect mode, causes OOM"GitHub issue

Strix is an autonomous AI penetration-testing agent that attacks your app like a real hacker and validates every finding with a working proof-of-concept before reporting it, aiming to eliminate the false positives that plague static scanners. It has climbed to strong open-source adoption with active releases and a commercial managed-pentesting platform behind it, and it recurs on 'top AI pentest tools' roundups - though users flag meaningful token cost and residual model hallucination as caveats.

• ~35,500 GitHub stars, 3,600 forks, 16 releases (latest v1.0.4, June 2026)
• Show HN reached 102 points and 15 comments
• Product Hunt launch rated 4.5/5 with 236 followers
• Commercial platform at strix.ai offers SSO, compliance reports, and Jira/Linear integrations
• Covered by Help Net Security plus multiple independent hands-on reviews

• "it feels like having an AI pentester built into your workflow - one of the most impressive dev tools I've used"Product Hunt comment

• "reports took about an hour and came back pretty complete and accurate"Product Hunt comment

• "communication is disastrous - waited almost a week for Discord approval"Product Hunt comment

• "the multi-agent architecture isn't just marketing; it mirrors how real pentest teams work"Medium article

• "a comprehensive assessment burned close to 60 million tokens over ~12 hours"Dev.to article

• "LLMs can still hallucinate a vuln because the code looks like a vulnerable pattern - verify"Dev.to article

• "on macOS it reports 'DOCKER NOT AVAILABLE' and won't start even though docker works"GitHub issue

An official OpenAI plugin that lets developers invoke OpenAI's Codex from inside Anthropic's Claude Code - running code reviews, adversarial reviews, and delegated 'rescue' tasks without leaving the session. Its real significance is as a rare cross-vendor interop tool: two competing AI-coding vendors' agents working in the same terminal, enabling a 'different model reviews the code another model wrote' loop that reduces single-model blind spots.

• ~23,700 GitHub stars, Apache-2.0, shipped under the official openai org
• 6 releases, latest v1.0.5 on June 23, 2026; ~145 issues and 123 PRs signal live iteration
• Ships 7 slash commands including review, adversarial-review, rescue and transfer
• Broad third-party coverage within weeks - OpenAI dev community, 10+ YouTube tutorials, numerous write-ups
• Roughly 30-second, four-command install; free ChatGPT tier works as the Codex backend

• "I'm surprised this isn't getting more attention"OpenAI community

• "one of the most-clicked links I have ever shared here"OpenAI community

• "the code is normal, so to speak, very well designed"OpenAI community

• "Codex found 4 high-severity issues and Opus found 8, with only 1 overlap"Dev.to article

• "no switching, no copy-pasting context - the review happens where the code was written"Dev.to article

• "said 'yes, fix everything' once and came back to a diff full of renamed variables"Dev.to article

• "the review gate can create long-running loops and burn usage quickly"Dev.to article

An official Alibaba open-source project that flips browser automation inside-out: instead of driving a browser from an outside process, it embeds a natural-language GUI agent directly into a web page as plain JavaScript, reading the live DOM as text with no extension, headless browser, or screenshots. It has captured strong developer momentum and a bring-your-own-LLM design, making it a natural fit for dropping AI copilots into existing SaaS, ERP, and CRM interfaces.

• ~22,800 GitHub stars, ~2,000 forks (MIT license) on the official alibaba/page-agent repo
• 34 releases with an active cadence, latest v1.11.0 on July 3, 2026
• ~46,800 npm downloads in the trailing month
• Show HN reached 147 points and 76 comments
• Broad LLM support (OpenAI, Claude, Qwen, Gemini, DeepSeek, Grok, Ollama) plus a Chrome Web Store extension

• "i tested the chrome extension, it worked great - navigated settings, clicked tabs, scrolled and found it"HN comment

• "the 'inside-out' framing resonates - I build embeddable scripts dropped into third-party sites"HN comment

• "the biggest challenge is the tension between deep DOM access and maintaining isolation"HN comment

• "bookmarklets are such an underrated feature for injecting and testing scripts"HN comment

• "it scrolled to the bottom of the page without any action on my part"Review article

• "instant 'task aborted' error when executing tasks via MCP due to a React re-render"GitHub issue

• "unable to use custom large language models"GitHub issue

Glaze is a new AI app-generation tool from Raycast: describe an app in plain language and it builds a real native Mac app that lives in your dock, launches instantly, and works offline. Coming from an established, venture-backed team and being integrated directly into the Raycast launcher, it carries unusually strong company signals - though a vocal minority questions trust and permissions for AI-generated software and whether it goes beyond happy-path demos.

• #1 Product of the Day on Product Hunt with ~555 upvotes
• 234-point, 145-comment debate on Hacker News with active founder participation
• Built by Raycast, which raised a $30M Series B led by Atomico (Accel, Coatue, YC participating)
• Shipping as a built-in Raycast feature plus a public store for community-built apps
• Built on Claude Code and OpenAI Codex under the hood; Raycast's own support team runs on Glaze-built apps

• "super fun - kind of the dream of site-specific browsers brought to the generative AI era"Product Hunt comment

• "software adapting to the individual instead of forcing everyone into the same workflow really resonates"Product Hunt comment

• "you get a working app one-shotted pretty much all the time; our support team runs on Glaze apps"HN comment

• "it springboards off Raycast's teams feature so well it gives it a real reason to exist"HN comment

• "a highly specialized tool for desktop that actually works"HN comment

• "I cannot bring myself to trust unreviewed software with arbitrary permissions on my machine"HN comment

• "my metric for this: did Glaze build the Glaze app?"HN comment