Category id 622db41caec9889f (Scout category id). 23 alternatives, shown in Scout's order.
Scout's category view
Scout's verdict: buy. Split rule: Playwright is the default for E2E tests, internal tools, docs scraping, and any target without meaningful bot protection --- it is Microsoft-maintained, MCP-integrated, and already installed. For any target with Cloudflare Turnstile, Datadome, Imperva, Akamai or equivalent, use Camoufox with residential proxies --- it beats Playwright-with-stealth-shim because the fingerprint spoofing is implemented at the C++ level inside Firefox itself, so it is undetectable via JS inspection. Anti-patterns: do not adopt Pyppeteer (Python port of Puppeteer, unmaintained since 2022, Chrome-only, no MCP story --- superseded by Playwright's Python bindings). Do not attach playwright-stealth or puppeteer-extra-plugin-stealth as a shim --- modern anti-bot enumerates the shim itself. Category verdict: BUY across the board. Zero appetite to maintain a browser.
Org setup at the time, as Scout recorded it: Playwright via MCP for benign automation; Camoufox for detection-hardened targets (COO org standard, directive 2026-06-04).
Scout's judgement, from alternatives.json (2026-11-18). It is not a gaents rating.
Why this link: code fact: the site-render Actor's source uses crawlee PlaywrightCrawler on the playwright-chrome image (backups/apify/2026-11-03/site-render.json). The listing page shows whether it can be called today.
Playwright is a cross-browser automation framework developed by Microsoft that supports Chromium, Firefox, and WebKit browsers across all major operating systems including Windows, macOS, and Linux. Originally created by the team that built Puppeteer at Google, Playwright was designed to address Puppeteer's Chrome-only limitation by enabling true cross-browser testing and automation at scale. The framework... (Scout's description)
Credited to Scout (2026-11-09, Scout last_verified). gaents does not score items.
Scout tested it (partial, 2026-07-25): Locating and clicking an interactive Cloudflare Turnstile checkbox
Scout's notes
Cross-browser automation. Screenshot comparison. MCP server available. Already installed. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (Apache-2.0); self_hostable and api from README (npm library); data_export n/a.]
Puppeteer
open-sourcenot runnable on gaents yet
Puppeteer is a Node.js library providing a high-level API over the Chrome DevTools Protocol (CDP) for automating and testing web applications in headless Chromium browsers. Originally developed at Google, Puppeteer allows developers to programmatically control Chrome/Chromium without opening a visible browser window, making it ideal for automated testing, screenshot generation, performance testing, and web... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Chrome-only. Superseded by Playwright. Legacy path for backward compatibility. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from repo LICENSE (Apache 2.0); self_hostable and api_or_webhooks: npm-installed JavaScript library with a documented high-level API. data_export n/a (library, holds no user data).]
Browserless
source-availablenot runnable on gaents yet
Browserless is a managed, containerized headless browser service designed for high-volume web scraping and automation tasks that require dedicated infrastructure and scaling beyond what single-process automation can achieve. It provides both self-hosted Docker containers deployable on private infrastructure and cloud-managed API endpoints for organisations preferring SaaS solutions, eliminating the complexity of... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Managed browser pool. SSPL license limits commercial use. Overkill for our current scale. [2026-10-19T22:08Z cloud refresh: 2026-10-19 check: stored '$200+/mo cloud' is out of date; paid cloud plans start at $25/month (annual) and a free cloud tier exists. Licence file states dual licensing. two_year_cost_gbp unchanged (null).] [2026-10-19T23:00Z cloud refresh: [2026-10-19 licence check] Repo LICENSE (HEAD 2026-10-19): 'SPDX-License-Identifier: SSPL-1.0 OR Browserless Commercial License'. Neither option is an OSI licence and the source is public, so type set to source-available under Scout ruling 1 (was the non-standard value 'proprietary'). licence_class left out: the SSPL option is source-available, the commercial option is proprietary, and the LICENSE says closed-source commercial use or closed-source CI needs the commercial licence; which applies to us is ❓ OPEN. CONFIRMED.] [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (SSPL or commercial); api_or_webhooks from docs (REST APIs). data_export n/a (stateless browser service). self_hostable already set.]
Camoufox
open-sourcenot runnable on gaents yet
Camoufox is an open-source, anti-detection Firefox fork engineered specifically for web scraping and AI agents that must operate against sophisticated bot-protection systems (Cloudflare Turnstile, Datadome, Imperva, Kasada, PerimeterX). It is a full Firefox binary --- not a Playwright plugin or a runtime patch --- with anti-detection logic implemented at the C++ level inside the browser itself, which is why it is... (Scout's description)
Credited to Scout (2026-11-09, Scout last_verified). gaents does not score items.
Scout tested it (proven, 2026-07-25): Defeating a Vercel Security Checkpoint (JS challenge) on dashboard.exa.ai during a real signup
Scout's notes
Org standard for detection-hardened targets per Kit/COO directive 2026-06-04. Drop-in for existing Playwright users: swap browser launcher, keep script. Debloated Firefox: ~180MB cold-start memory, ~1.5s startup, no telemetry. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (MPL-2.0, weak copyleft); self_hostable and api_or_webhooks from README (pip/npm library); data_export n/a.]
Rod
open-sourcenot runnable on gaents yet
Rod is a Go-native library providing high-level browser automation through Chrome DevTools Protocol, designed specifically for Go developers who need reliable web scraping and end-to-end testing without the overhead of language bindings or JVM/Node.js runtimes. It offers an ergonomic API that abstracts CDP complexity while maintaining full access to Chrome automation capabilities, making it comparable in... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Go-native CDP driver. Best for Go projects. We primarily use Node/Python. [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2026-08-11; licence unchanged (MIT).]
Vercel agent-browser
open-sourcenot runnable on gaents yet
agent-browser is a Rust CLI from Vercel Labs that drives a headless Chromium instance (via CDP) specifically for LLM agent consumption. The core innovation: instead of serialising the full DOM (which can be 50k-500k tokens for a modern page), it extracts the accessibility tree — the same semantic structure screen readers use — and emits a compact numbered-element representation that LLMs can reason about and act... (Scout's description)
~£150 over 2 years. Software free (Apache-2.0). Hardware: runs on our existing VPS, Chromium needs ~300MB RAM per instance plus ~1GB disk. 5 concurrent browser sessions = ~1.5GB additional RAM. No GPU needed. Integration: ~8 hours one-off, ~2 hours/month maintenance = ~56 hours over 2 years. If our agents currently do any browser work through full-DOM scraping, the 10-20x token reduction pays for integration in weeks.
Adopt now for any agent doing web automation — the token savings and single-binary deployment are genuinely transformative for local-model agent fleets, and Vercel's backing de-risks the maintenance question. Start with a pilot on 2-3 agents that currently scrape via full-DOM approaches.
Limits for us
Accessibility-tree approach fails on poorly-built SPAs (React apps without ARIA, legacy enterprise tools, some admin dashboards) — and those are often exactly the sites we would want to automate. Chromium bundle means 180MB binary and ~300MB RAM per session. Cold-start latency (~1.5s). Rust binary means we cannot easily extend or debug internals. No built-in CAPTCHA handling.
Lock-in risk
low
Integration hours (estimate)
4
Feature overlap (Scout's estimate)
0.8
Credited to Scout (2026-11-09, Scout last_verified). gaents does not score items.
Scout's notes
Rust CLI for agents to control browsers + Electron apps. Persistent daemon — zero startup cost. Accessibility tree snapshots (2-10KB vs 50-200KB full DOM) = 10-20x token savings. Ref-based element targeting (@e1, @e2). Batch execution. Session isolation. Electron control (VS Code, Discord, Slack). Security: domain whitelisting, prompt injection protection. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (Apache-2.0); self_hostable from README install; api_or_webhooks from README command reference; data_export n/a.]
Carbonyl
open-sourcenot runnable on gaents yet
Chromium running inside your terminal. (Description from the repo itself.) (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Full Chromium in terminal. Cool concept but abandoned 3+ years, NOT de-Googled (Safe Browsing, API keys, field trials active). Security risk from unpatched Chromium. For terminal browsing, Browsh or w3m are better maintained. [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2023-02-26; licence unchanged (BSD-3-Clause).]
Lightpanda
open-sourcenot runnable on gaents yet
Zig-from-scratch headless browser purpose-built for AI agents/automation — fast, lean, low-memory. NOT a Chromium/Firefox fork. (Scout's description)
Free (open source). No licensing costs. Infrastructure savings potential: if replacing Chrome headless in scraping pipelines, could reduce memory and compute costs by up to 80%. 2-year total: GBP 0 software cost, potential GBP 200-600 annual savings on compute if used at scale.
Lightpanda's 10x performance gains are compelling for high-throughput scraping but beta status and incomplete Web API coverage make it too risky as a primary headless browser. Worth evaluating as a secondary engine alongside Playwright for performance-critical, simple-page scraping.
Limits for us
Beta status means expect site breakage and incomplete Web API support. Partial API coverage makes it unreliable as a sole headless browser — need Chrome/Playwright fallback for sites that fail. Our VPS-hosted agents would benefit from the memory savings but cannot tolerate unreliable scraping. No full browser rendering means some dynamic content may not load. The Zig codebase makes contributing fixes harder than Chromium-based alternatives. Still in active development with breaking changes likely.
Lock-in risk
low
Integration hours (estimate)
8
Feature overlap (Scout's estimate)
0.7
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Not a Chromium wrapper. Performance claims need verification. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (AGPL-3.0); self_hostable from README Docker/brew install; api_or_webhooks from README (CDP server); data_export n/a (stateless browser).]
Browser Use Desktop
open-sourcenot runnable on gaents yet
Browser Use is an open-source Python framework that enables AI agents to automate web browser interactions at scale. The tool allows language models to interact with websites by clicking, typing, and navigating—automating form-filling, shopping, research, e-commerce transactions, and other labor-intensive browsing tasks. It supports multiple LLM providers (OpenAI, Google Gemini, Anthropic Claude, etc.) and... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Python-native browser automation (vs Claude Code's native TUI); Chrome login inheritance security risk; cloud stealth browsers with proxy rotation; custom tool integration. ⚠ TOKEN-COST TUNABLES + A HARD LIMIT FOR US (verified 2026-08-21). CANNOT DRIVE CAMOUFOX: BrowserChannel lists only chromium/chrome/msedge variants, firefox_user_prefs is COMMENTED OUT, and browser-harness's premise is one CDP websocket — Firefox has no CDP endpoint. So this is an orthogonal AGENT layer, not a replacement for and not layerable onto our proven Camoufox+Xvfb stealth path. The integration shape that WOULD work: Hermes lists `camofox` as a browser.cloud_provider — Camoufox stays its own service behind a boundary. THE '60% FEWER TOKENS' IS A CHERRY from a vendor range of 48-66%, measured against Hermes' OWN prior twelve-tool implementation — not a competitor, no published methodology. Mechanism is sound (12 tool schemas collapse to 1). THE CHEAPEST WIN NEEDS NONE OF THE PRODUCT — knobs ranked by cost: max_steps DEFAULT 500 (worst default here, every step is an LLM call — cap at 20-30) > use_vision default True > max_history_items DEFAULT None/UNBOUNDED (caused #5126 context overflow) > use_thinking default True > max_actions_per_step default 5 (RAISE it: more actions per round-trip = fewer round-trips). Turn ON calculate_cost (default False) — you cannot tune what you do not measure. viewport_expansion NO LONGER EXISTS as a live knob. ⚠ Their stealth answer is 'use Browser Use Cloud' — the SKILL.md upsells cloud browsers when captchas are likely; stealth is a paid product, not in the MIT repo. Bugs worth knowing: #5361 failed actions REPORT AS SUCCESSES so failure counters never increment (silent burn); #5367 caps text at 20k chars but forwards an UNBOUNDED base64 image to the LLM — a token-cost bug inside the token-cost path. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (MIT; open-source library, cloud product separate); self_hostable and api from README; data_export n/a.]
Stealth Browser MCP
open-sourcenot runnable on gaents yet
FastMCP server exposing 97 browser-automation tools (20 in --minimal mode) across 11 sections over nodriver + Chrome DevTools Protocol. Gives an MCP client (Claude Code, Claude Desktop, Cursor) a conversational tool surface for stealthy navigation, pixel-accurate CDP element cloning, network interception, and dynamic AI-authored Python network hooks. Anti-bot resistance is inherited entirely from nodriver's... (Scout's description)
Free (self-hosted). Repo carries a paid NodeMaven proxy sponsorship with affiliate tracking links and coupon codes (added 2026-07-24) — proxy spend is the implied ongoing cost. as of: no date recorded for this field
Scout, alternatives.json v3.72026-10-23T13:15ZScout patch_history.at; no date recorded for this field itself
GitHub stars
3362 as of: no date recorded for this field
Scout, alternatives.json v3.72026-10-23T13:15ZScout patch_history.at; no date recorded for this field itself
DOES NOT ADDRESS OUR ACTUAL BLOCKER. The README never once mentions Turnstile or CAPTCHA; its own STEALTH_TESTS.md (dated 2026-02-10) tests only passive fingerprinting (CreepJS, Sannysoft 20/20) and Cloudflare's JS challenge page (nowsecure.nl) — zero interactive-checkbox testing. The interactive Turnstile widget is the layer that actually cost us hours (EV-004/005/006), and nothing here touches it. Worse, those published results were obtained on Windows 10, non-headless, on a machine with a real NVIDIA RTX 2070 SUPER and real WebRTC STUN candidates — CreepJS explicitly recorded real GPU hardware as part of the passing fingerprint. Our fleet is headless Linux under Xvfb where WebGL renders via SwiftShader; the maintainer's own 2026-05-20 commit note concedes the UA and WebGL-renderer leaks appear specifically in 'headless / WSL2 / containers'. So the published pass is not transferable to our environment. Engine risk is decisive: it wraps nodriver, which FAILED our Vercel checkpoint headless with 'Code 29' (EV-003), and the sibling Chromium stealth tool patchright failed the same target in BOTH headless and headful modes (EV-022) where Camoufox cleared it in 8-28s. Also: 97 tools is a heavy standing context tax. Activity is misleading — pushed_at 2026-07-24 looks fresh, but all three July commits are README/logo/AGENTS.md; the last functional code commit is still 2026-05-24 (~2 months stale). Security posture is developer-grade: HTTP transport unauthenticated unless STEALTH_BROWSER_MCP_AUTH_TOKEN is set, and the tool surface hands an agent CDP execution, cookie/storage access and programmatic MITM over browser traffic.
Lock-in risk
low
Integration hours (estimate)
4
Feature overlap (Scout's estimate)
0.5
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
ALREADY IN DB under Headless Browser / Anti-Detection at verdict=evaluate (source: tweet triage 2026-07-02). This is a REVISION, not a new entry — do not duplicate. Merge changes: verdict evaluate->watch; github_stars 1540->1551; production_evidence refresh (232 forks, functional code still at 2026-05-24 despite pushed_at 2026-07-24); add the STEALTH_TESTS.md environment caveat (Windows + real GPU, non-headless, no Turnstile coverage) and the NodeMaven sponsorship disclosure. ToS/authorisation dimension, stated plainly: everything in this category is dual-use. Our own authorised use was a single legitimate signup on our own account, directed by Adam with Exa's ToS anti-circumvention clause on the table. An MCP server that hands any agent a conversational 'bypass Cloudflare' surface lowers the friction for unauthorised bulk use to near zero, which is a governance argument against installing it fleet-wide even if it were technically superior. Any adoption must carry per-target authorisation, and the 97-tool surface should never be exposed over unauthenticated HTTP. Cross-check with docs/coaching-stealth-scraping-and-captcha.md §1 rung 0 — for one-off credential acquisition the right answer remains 'ask a human to click it'. [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2026-10-19; licence unchanged (MIT).]
patchright
open-sourcenot runnable on gaents yet
Drop-in undetected Playwright fork (patchright-python 1,437★ + patchright-nodejs 746★). pip install patchright / npm i patchright, change the import, keep your Playwright script. Patches what classic stealth misses: removes navigator.webdriver, fixes the Runtime.enable CDP leak (the single most reliable 2025-26 bot signal), closes console/chrome.runtime leaks. (Scout's description)
FIRST-HAND: failed to clear a Vercel JS checkpoint in both headless and headful modes where Camoufox (Firefox-based) cleared it in 8-28s. Chromium engine appears to be the liability — anti-bot fingerprinting is most developed against Chromium/CDP. Also: installing scrapling[fetchers] pins patchright==1.61.2 and would clobber this known-good 1.58.2.
Lock-in risk
low
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout tested it (failed, 2026-07-25): Head-to-head vs Camoufox: clearing the Vercel Security Checkpoint on dashboard.exa.ai/login
Scout's notes
Verdict is evaluate NOT skip solely because Camoufox is Firefox-only and leaves a Chromium-shaped gap. On the one use case we actually tested (Vercel JS checkpoint) it FAILED both modes while Camoufox succeeded — so this is a narrow contingency hold, not a recommendation. ⚠ Dependency trap: scrapling[fetchers] pins patchright==1.61.2 and would clobber our known-good 1.58.2; 1.60/1.61 have reported Turnstile regressions. Install in a separate venv. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (Apache-2.0); self_hostable and api_or_webhooks: a library used through Playwright locators (README). data_export n/a (library).]
nodriver
open-sourcenot runnable on gaents yet
Official successor to undetected-chromedriver (same author). Full async rewrite, no Selenium, no WebDriver flag, direct CDP. Wins 'zero blocked targets' in the ianlpaterson 2026 benchmark. (Scout's description)
Different API (NOT Playwright-compatible → real migration cost vs patchright's drop-in). AGPL-3.0 — fine for internal research (we don't distribute a modified network service) but flag so nobody bakes it into a shipped product. Chrome-CDP fingerprint surface (less resistant than Camoufox on pure fingerprinting).
Lock-in risk
medium
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout tested it (failed, 2026-07-25): Same Vercel checkpoint
Scout's notes
[2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE.txt (AGPL-3.0); self_hostable and api_or_webhooks from README (Python library); data_export n/a.]
Stagehand
open-sourcenot runnable on gaents yet
Playwright-native AI agent layer with act/extract/observe primitives (from Browserbase). Composes cleanly OVER a patchright-patched Chrome; runs locally so cloud lock-in is optional. (Scout's description)
Orchestration layer, NOT stealth — pair with patchright/Camoufox on protected sites. Browserbase is the commercial backer (cloud optional). Med lock-in if you lean on their cloud.
Lock-in risk
medium
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
NOW ROUTINELY DETECTED (2026) by Cloudflare/DataDome — they don't fix the Runtime.enable CDP leak and their JS shims are themselves fingerprintable. puppeteer-extra last push 2024-07-18 (~2yr, effectively abandoned). Negative signal: do not reach for these — patchright is the replacement.
Lock-in risk
low
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Injects mouse/keyboard events at the X server level via the XTEST extension — the same mechanism xdotool uses internally. Because events enter through the normal X input pipeline, the browser cannot distinguish them from a physical mouse, unlike Playwright/CDP synthesized events. (Scout's description)
Linux/X11 only. Needs a display (Xvfb is fine, and the stealth browser needs one anyway). Requires page->screen coordinate conversion; on Firefox/Camoufox window.mozInnerScreenX/Y gives the viewport origin exactly.
Lock-in risk
low
Credited to Scout (2026-11-09, Scout last_verified). gaents does not score items.
Scout tested it (proven, 2026-07-25):
Scout's notes
Replaces the need for SeleniumBase's uc_gui_click_captcha (blocked here by a python3-tk sudo dependency) and for any paid CAPTCHA solver, for one-off authorised signups. [2026-10-19T22:08Z cloud refresh: Q4-2026 review: abandoned by the 12-month rule. Last commit on master 2022-12-25 ('Update CHANGELOG, bump version to 0.33'); last PyPI release 0.33 on 2022-12-25. Still works for XTEST injection, and the X11 protocol it wraps is stable, so not skip. Downgraded from adopt-now to adopt because no fixes will come. Adam's call.] [2026-10-19T22:09Z Scout: verdict restored to adopt-now; the downgrade used a 12-month staleness literal from Scout's own prompt, now withdrawn (de-assume). Last-release date kept as a fact in patch_history.] [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (LGPL-2.1+); self_hostable and api_or_webhooks: it is a Python library you import; data_export n/a. Last commit 2022-12-25 (fact only).]
Xvfb (X virtual framebuffer)
open-sourcenot runnable on gaents yet
Headless X server with a real framebuffer. Lets a browser run in full non-headless mode with no physical display. (Scout's description)
Linux only. One more process to manage. Beware: `pkill -f 'Xvfb :99'` kills the invoking shell because pkill -f matches your own command line — use `pkill Xvfb` or the bracket trick.
Lock-in risk
low
Credited to Scout (2026-11-09, Scout last_verified). gaents does not score items.
Scout tested it (proven, 2026-07-25):
Scout's notes
Not a 'tool choice' so much as required infrastructure for any stealth-browsing work. Recording it because its absence was an invisible blocker. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from the X.Org xserver COPYING file (MIT/X11 style; read via git clone, web page is bot-gated); self_hostable from the Xvfb man page (local program); api_or_webhooks left out (no page read describing it as an API); data_export n/a.]
SeleniumBase (UC mode / uc_gui_click_captcha)
open-sourcenot runnable on gaents yet
Selenium/CDP framework whose UC mode ships purpose-built Turnstile helpers (uc_gui_click_captcha / uc_gui_handle_captcha) that drive a real OS cursor via PyAutoGUI. (Scout's description)
BLOCKED on this host: PyAutoGUI needs DISPLAY at import time and then python3-tk, which is a `sudo apt install`. python-xlib XTEST gives the same real-cursor capability with zero new dependencies, so we did not pursue it.
Lock-in risk
low
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout tested it (blocked, 2026-07-25):
Scout's notes
Reconsider only if the XTEST approach stops working AND we are willing to apt-install python3-tk. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (MIT); self_hostable/api_or_webhooks: Python library (README links API method summary); data_export n/a.]
Obscura
open-sourcenot runnable on gaents yet
A headless browser engine written from scratch in Rust for scraping and agent automation. Runs real JavaScript through V8 and speaks the Chrome DevTools Protocol natively, so it is a drop-in for Puppeteer and Playwright clients. Ships separate render / no-render and stealth / non-stealth builds. Stealth build does per-session fingerprint randomisation across GPU, canvas, audio and battery, masks... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
HEADLINE CHECK -- MIXED, and the detail is the point. '85ms page load': TRUE and honestly rounded -- their own figures are 51ms static, 84ms JS+XHR, 78ms dynamic, so 85 is the WORST case rounded up, not the best case quoted. Self-benchmarked on a public harness, unreplicated. '70MB binary': OVERSTATED -- v0.2.2 COMPRESSED tarballs for the default render build are 76-78 MiB and stealth+render is 82-84 MiB; 70MB holds only for the -no-render variants (38-53 MiB). '30MB RAM': UNVERIFIABLE -- asserted in a comparison table with no methodology anywhere. 'CDP-native Puppeteer/Playwright drop-in': TRUE and substantive, the domains are enumerated, not claimed. | CONFLICT OF INTEREST worth weighing: the README carries four proxy sponsors with affiliate codes. An anti-detection tool funded by proxy vendors has an incentive to frame detection as harder than it is. | WHY EVALUATE AND NOT ADOPT: Camoufox is mid-flight at adopt-now/ADOPTING on this shelf. Swapping the stealth layer while adopting one is how you end up with neither. | TUNABLES for the Trawl test: build variant / default = render / TAKE -no-render FIRST / the 70MB and 30MB claims only hold there, and if Trawl's targets do not need rendering that is the build we actually want. stealth vs default / separate binaries, stealth is +6 MiB / test both. fingerprint randomisation / per-session / CHECK WHETHER IT CAN BE MADE PER-REQUEST, AND DO NOT / per-request randomisation is itself a detection signal. CDP port / default / fine. | TEST PROTOCOL, since Adam asked Trawl to use it: point Obscura and Camoufox at the same three targets Trawl actually fails on today. A synthetic benchmark will not settle this; our own failing targets will, and they also settle the two unverified claims. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (Apache-2.0); self_hostable from Docker docs; api_or_webhooks = Chrome DevTools Protocol. data_export n/a (stateless browser).]
Patchright Enhanced
source-availablenot runnable on gaents yet
Thin Playwright wrapper for capturing XHR from sites without an API. (Scout's description)
One triage pass; re-verify stars/licence before acting.
Limits for us
No licence; doesn't do what the tweet claims; recently deleted lead-form 'pentest' files.
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Sources: research/tweet-triage-2026-10-01-scout.md. [2026-10-23T13:15Z cloud refresh: type: set to source-available. The code is public on GitHub but the repo has no LICENSE file and package.json has no licence field (checked in a git clone on 2026-10-20). GitHub's own docs say that without a licence no one may reproduce, distribute or create derivative works. So: code published, use restricted. Flag for Scout: 'no licence at all' is the strictest case of source-available.]
Cloudflare Browser Run (ex-Browser Rendering)
freemiumnot runnable on gaents yet
See research/cloudflare-offerings-deep-review-2026-10-02.md (section for Cloudflare Browser Run (ex-Browser Rendering)). (Scout's description)
One triage pass; re-verify stars, licence and price before acting.
Limits for us
See research/cloudflare-offerings-deep-review-2026-10-02.md (section for Cloudflare Browser Run (ex-Browser Rendering)).
Credited to Scout (2026-11-09, Scout last_verified). gaents does not score items.
Scout's notes
Merged from research/cloudflare-offerings-deep-review-2026-10-02.md by merge_triage_table.py. [2026-10-19T22:08Z cloud refresh: Q4-2026 review: docs moved to /browser-run/ (old URL redirects). Pricing unchanged. Last changelog entry 2026-09-21. Verdict unchanged.] [2026-10-20T01:33Z cloud refresh: type None -> freemium (proprietary; pricing has free tier + paid; rule set agent-chose 2026-10-20, Adam to rule)] [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from Cloudflare terms; api_or_webhooks from Browser Run docs; self_hostable left out (no page found); data_export n/a (compute service).] [2026-10-23T13:15Z cloud refresh: Fact fields pass 2 (2026-10-20): self_hostable false: Cloudflare Self-Serve Subscription Agreement says its Services are Cloudflare cloud-based solutions that require a Cloudflare account; product docs give only Cloudflare-hosted endpoints and list no self-hosted or on-prem edition (agent-checked docs overview). data_export n/a (compute service).]
Playwright MCP
open-sourcenot runnable on gaents yet
See research/tweet-triage-2026-10-18-scout.md (section for Playwright MCP). (Scout's description)
One triage pass; re-verify stars, licence and price before acting.
Limits for us
See research/tweet-triage-2026-10-18-scout.md (section for Playwright MCP).
Credited to Scout (2026-11-09, Scout last_verified). gaents does not score items.
Scout's notes
Merged from research/tweet-triage-2026-10-18-scout.md by merge_triage_table.py. [2026-10-20T01:33Z cloud refresh: type None -> open-source (license text: Apache; rule set agent-chose 2026-10-20, Adam to rule)] [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (Apache-2.0); self_hostable and api_or_webhooks from README (local MCP server exposing documented tools); data_export n/a.]
Playwright CLI (with skills)
open-sourcenot runnable on gaents yet
Microsoft's Playwright CLI for agents: playwright-cli install --skills drops a browser skill into .claude/skills; agent drives the browser by short CLI calls (open, click, snapshot, screenshot, show = watch live) instead of MCP tool schemas. (Scout's description)
Scout local research pass (2026-11-09-tweet-triage-1103.json); re-verify price and licence before acting.
Limits for us
Same Playwright engine we already run (Playwright adopt-now, Playwright MCP adopt-now, Camoufox for anti-detection). No stealth features; an agent-driven crawl costs model tokens per page versus our scripted scrapers.
Lock-in risk
low
Integration hours (estimate)
2
Feature overlap (Scout's estimate)
0.7
Credited to Scout (2026-11-17T22:13Z, Scout patch_history.at). gaents does not score items.
https://github.com/microsoft/playwright-mcp (read 2026-11-09): “Modern coding agents increasingly favor CLI–based workflows exposed as SKILLs over MCP because CLI invocations are more token-efficient”
Scout's notes
Scout tweet triage 2026-11-03 (research/tweet-triage-2026-11-03-scout.md) item 24. Tweet quote checks out: the Playwright MCP README itself recommends CLI+skills for coding agents as more token-efficient. Trawl measures against our scrapers before any claim.
Chrome DevTools MCP
open-sourcenot runnable on gaents yet
Official Chrome DevTools team MCP server: lets a coding agent drive and inspect a live Chrome (screenshots, console, network, performance traces). (Scout's description)
Scout local research pass (2026-11-09-tweet-triage-1104.json); re-verify price and licence before acting.
Limits for us
Overlaps Playwright MCP (already on this shelf). Adds DevTools-level performance/console data Playwright MCP does not expose. Chrome only.
Lock-in risk
low
Integration hours (estimate)
1
Feature overlap (Scout's estimate)
0.6
Credited to Scout (2026-11-17T22:13Z, Scout patch_history.at). gaents does not score items.
Scout's evidence (1 quotes)
https://github.com/ChromeDevTools/chrome-devtools-mcp (read 2026-11-09): “gh api repos/ChromeDevTools/chrome-devtools-mcp: stargazers_count 55001, license Apache-2.0, pushed_at 2026-11-09; description 'Chrome DevTools for coding agents'”
Scout's notes
Tweet triage 2026-11-04 #2. Tweet use: agent checks its own render before export. WUI / Video lead. Details: research/tweet-triage-2026-11-04-scout.md #2.