ChromeWalk 14.0.0

Smarter captures, a far lighter MCP footprint, and one-line verified installs on Windows and macOS.

Highlights

• Lean MCP. The tool list your AI assistant loads is a fraction of its former size, the server answers in well under a second and does nothing while idle. Every option is still there: rare ones sit under advanced, and any tool answers help: "brief" or help: "full" without running. Large results come back as a short summary plus the path of the full file.
• Readable file names and a navigation log. Capture tools name files after the host, path, page title and viewport, and write a nav-log.jsonl that records where every file came from. --no-smart-names and --no-nav-log restore the old behaviour.
• Content metrics in every audit and design run: words, reading time, readability, images without alt text, headings, title and description length. Opt-in prose comparison (compare --prose) and cross-page de-dupe (audit --dedupe).
• SEO layer for audit (--layers seo, --check-links): titles, descriptions, canonicals, structured data, broken links and redirect chains, from the same captures.
• One walk collector for every walking tool, with --walk-profile scrape|audit|parity, plus an engine parity matrix for Chromium, Firefox and WebKit.
• Polite pacing, always on and adaptive per host: free until a site signals 429, 503, Retry-After or a challenge. --pacing off for your own sites.
• Session liveness: --interact, --keep-alive and stale or expired session detection.
• Network options: a generic --proxy on every browser tool, and ChromeWalk works with Nimble (opt-in, off by default).
• Zero leftover processes. A run that is cancelled, fails or is killed leaves no browser or driver behind, on every engine.
• Verified install scripts and a macOS package. install.ps1 and install.sh check every download against SHA256SUMS.txt before installing anything, install per user without admin rights and run setup. The macOS package is for Apple Silicon and is not yet signed (System Settings > Privacy & Security > Open Anyway, or use install.sh, which avoids the prompt).

Behaviour changes you will notice

• Capture file names follow the new pattern (named flow steps and render baselines keep their names).
• Content counts are on in audit and design (audit --no-content-metrics turns them off).
• schema counts top-level entities only; nested ones are listed and counted under counts.nested.
• Polite pacing is always on, and the walk collector adds a walk section to results and one summary line on stderr. stdout stays one JSON document.
• --interact is on by default in the audit and parity walk profiles.
• schema --engine and cms --engine now render in the engine you name.
• MCP tools/list is tiered; options passed at the top level keep working.

Install

Windows (PowerShell):

powershell -c "irm https://github.com/gator8125/ChromeWalk-MCP/releases/latest/download/install.ps1 | iex"

macOS (Apple Silicon):

curl -fsSL https://github.com/gator8125/ChromeWalk-MCP/releases/latest/download/install.sh | bash

Or download ChromeWalk-Setup.exe (Windows installer), the Claude Desktop extension chromewalk-14.0.0-win32-x64.mcpb, or ChromeWalk-14.0.0-macos-arm64.pkg below. Every asset's SHA-256 is in SHA256SUMS.txt.

ChromeWalk requires a license; chromewalk register <email> (or the chromewalk_register tool in your AI assistant) sends one.

ChromeWalk 13.0.0

ChromeWalk 13.0.0

Test the flows, not just the pages. 35 MCP tools (was 33).

Highlights

• Sign up and activate from your AI assistant. Two new tools, chromewalk_register and
  chromewalk_activate, work without a license: when a tool reports that a license is needed, the
  assistant can email you a key and activate the install inside the conversation. The key is never
  echoed back.
• Click-flow coverage. Each scripted smoke step can carry an expect (selector, text, URL change,
  absence) and is checked for console errors and failed requests it caused. --explore N clicks the
  safe controls you never scripted and never touches delete, log out, pay, submit or similar.
  --coverage reports how many controls were scripted, explored or skipped.
• Logged-in testing on every engine. xengine and render accept a warmed --session, --auth
  headers and a local-storage seed, so Chromium, Firefox and WebKit all test the signed-in pages.
• Edge and Brave are drivable browsers (--browser edge|brave|chrome|cft|ungoogled|<path>), always
  with ChromeWalk's own isolated profile.
• Setup completes the runtime. cw setup installs the browser engines and ends with a doctor
  check.
• Secure key handling. cw activate <key> --remember stores the key per Windows user; an install
  that can find a key activates itself once.
• No leftover Chrome. A killed run now ends its browser tree.
• Large design and audit runs run in batches and resume with --resume.
• Six more AI tools for chromewalk connect (16 in total): GitHub Copilot CLI, Kiro, Amazon Q
  Developer, Kilo Code, Goose and Visual Studio. An unparseable client config is never overwritten;
  connect --dry-run shows the plan.
• Agent skills: four ready-made skills (release gate, cross-engine, onboarding, troubleshooting)
  in skills/ of this repository.

Behaviour changes

• A failed smoke or xengine step now fails the unit (exit 3). Before 13.0 it was only logged.
• An unknown CHROMEWALK_BROWSER value is an error (exit 5) instead of being ignored; a misspelt
  --browser is a usage error (exit 2).
• cw procs only lists; cw procs --reap reaps.
• doctor adds two advisory checks (install_roots, startup_time) and takes a few seconds longer.
• Secret-named fields in API output are now also masked inside JSON strings and XML elements.

Assets

• ChromeWalk-Setup.exe: Windows installer
• chromewalk-13.0.0-win32-x64.mcpb: Claude Desktop extension
• SHA256SUMS.txt

ChromeWalk 12.0.0

ChromeWalk 12.0.0

Feedback-driven polish and automatic updates. 33 MCP tools.

Highlights

• Fewer false smoke failures. The browser fallback no longer dies with "Access is denied", and
  "StopIteration" crashes are now reported as infrastructure retries instead of failed pages. In a
  field pilot, 12.0.0 passed a real 48-page gate 48/0 in about half the time 11.0.1 took (46/48).
• Smoke that explains itself: every finding has a detail. Slow pages are told apart from failed
  ones, with per-page budgets learned on your machine. Failure artifacts are saved per page.
• Changed-pages-only runs: smoke --changed / --git-diff with a suite route_map.
• UI-uniformity toolkit: audit (now also audit <url> ...) and baseline.
• Run lifecycle: every run has a job file, a heartbeat line on long runs, and a clean stop if
  the calling command is killed. cw procs lists recent runs. Very large runs get a size warning.
• Pre-capture hooks and browser options on every capture tool.
• Automatic updates for the Windows app. Updates are signature-checked, checksum-checked,
  tested with a real page load and rolled back on failure. cw update --defer N and
  cw update --status control and show them.

Privacy, telemetry and updates

ChromeWalk 12 comes with updated Terms and
Privacy notices. Usage analytics are part of the license
and include the target URL (scheme, host and path; credentials, query strings and fragments are
removed on your computer first). cw telemetry status shows every field sent. Installs of 11.x
keep their current behaviour until they update.

Assets

• ChromeWalk-Setup.exe: Windows installer
• chromewalk-12.0.0-win32-x64.mcpb: Claude Desktop extension
• SHA256SUMS.txt

ChromeWalk 11.0.1

ChromeWalk 11.0.1 - hotfix (recommended for all 11.0.0 users)

Fixes pages that never load: smoke reporting "timeout: document never became interactive", and fetch ending on about:blank. This was most visible when the ChromeWalk data folder is on a slower (HDD) disk.

• Navigations that Chrome cancels while a fresh browser profile is still starting up are now retried. Previously 11.0.0 waited out the whole timeout.
• Removed a redundant cookie-clear step that stalled each launch by up to 30 s.
• fetch no longer crashes when printing page text that the Windows console can't display.

Upgrade: install the new .mcpb (Claude Desktop) or ChromeWalk-Setup.exe over 11.0.0. Your license carries over.

ChromeWalk 11.0.0

ChromeWalk 11.0.0

Design drift, self-healing, and probe-first runs. ChromeWalk runs real, local browser automation
for your AI agents, exposed as a local MCP server (chromewalk mcp, 31 tools) and a CLI. Everything
runs on your machine.

A license is required. Sign up at https://api.chromewalk.com/register.html, or run
chromewalk register you@example.com after installing, then chromewalk activate <key>.
Upgrading from 4.0, 6.1 or 9.0: activate before you switch. 10.0 and later fail closed without
a license.

Downloads

File	For
ChromeWalk-Setup.exe	Windows 10/11 x64. Installs chromewalk for every AI app (Claude Code, Codex, Cursor, VS Code, Gemini CLI, ...)
chromewalk-11.0.0-win32-x64.mcpb	Claude Desktop extension (Windows x64). Double-click to install

SHA256SUMS.txt lists both checksums.

The installer is not code-signed yet (the certificate is pending), so SmartScreen may warn.
Choose More info -> Run anyway, or verify the SHA-256 first.

What's new in 11.0

• UI/UX and design drift (design tool + compare layers):
  • tokens: palette with perceptual colour clustering, type scale, spacing grid, radii, shadows,
    breakpoints and CSS custom properties.
  • components: button, card, nav and other variant clusters, plus cross-page consistency.
  • css: rule-level CSS diff, custom properties, breakpoints, specificity and unused CSS.
  • ux: WCAG contrast, tap targets, focus visibility, overlap, layout shift and the call to
    action above the fold.
  • perceptual: SSIM visual similarity.
  • Hover/focus states and dark mode are covered. There is a design-system file import (DTCG /
    Style Dictionary), baselines, an HTML report, and a smoke page.design check.
• Probe-first, target-aware runs: probe checks DNS, TLS, redirects, bot protection, login
  walls and SPA vs server-rendered before any browser starts. Per-host target memory learns the best
  wait strategy and timeout, with one adaptive retry.
• Self-healing:
  • launch heal: reap, clear stale locks, quarantine a corrupt profile, fall back to another engine;
  • a spawn monitor that tracks every process ChromeWalk starts and reaps orphans and zombie runs,
    verifying identity before every kill;
  • an MCP watchdog;
  • chromewalk reset --soft/--hard.
• New MCP tools: design, probe, targets, reset, recommend (file a gap report from
  any agent).
• Reliability fixes from real agent use:
  • screenshots of local .html files render;
  • Playwright device names ("iPhone 13") resolve;
  • unknown MCP arguments are refused with a hint;
  • secrets (localStorage seeds, tokens, inline login flows) never appear on a command line;
  • the reCAPTCHA-footer false "blocked" is fixed;
  • logs --failures works for CLI runs;
  • doctor no longer fails a gate over an advisory check.
• Complete docs: a page per command (every flag, examples, an MCP call, troubleshooting),
  CONFIG, SCENARIOS (12 recipes), MCP client setup, exit codes and errors.
  chromewalk help <tool>, help --search.

Install

1. Run ChromeWalk-Setup.exe (per-user, no admin needed) and enter or request your license key.
2. Run chromewalk connect to register ChromeWalk in the AI apps on this machine.
3. Claude Desktop: open the .mcpb instead, and paste your key into the extension's settings.
4. Run chromewalk doctor to check that everything is healthy.

ChromeWalk 10.0.0

ChromeWalk 10.0.0

Reliability release. ChromeWalk runs real, local browser automation for your AI agents, exposed
as a local MCP server (chromewalk mcp, 26 tools) and a CLI. Everything runs on your machine.

A license is required. Sign up at https://api.chromewalk.com/register.html (your key arrives by
email), or run chromewalk register you@example.com after installing, then
chromewalk activate <key>.

Downloads

File	For	SHA-256
ChromeWalk-Setup.exe	Windows 10/11 x64. Installs chromewalk for every AI app (Claude Code, Codex, Cursor, VS Code, Gemini CLI, ...)	see SHA256SUMS.txt
chromewalk-10.0.0-win32-x64.mcpb	Claude Desktop extension (Windows x64). Double-click to install	see SHA256SUMS.txt

SHA256SUMS.txt lists both checksums. macOS packages are in testing and not published yet.

The Windows installer is not code-signed yet (the certificate is pending), so SmartScreen may show
"Windows protected your PC". Choose More info -> Run anyway, or verify the SHA-256 first.

What's new in 10.0

Multi-agent reliability. MCP calls now run concurrently, and each call runs in its own process.
A timeout or cancel cleans up the call's whole browser process tree.

Scenario, measured on the same machine	9.0	10.0
8 parallel tool calls on one server	fully serial, p50 24.4 s, max 47 s	overlapping, p50 8.9 s, max 15 s
Forced tool timeout	server hangs, 10-14 browser processes left behind	clean error in 5-10 s, 0 left behind
A hanging call beside other calls	blocks every call behind it	isolated, and it can be cancelled
4-agent smoke run on a saturated CPU	3 launch failures ("timed out during opening handshake")	0 launch failures

• Browser launches:
  • launches are staggered across every ChromeWalk process on the machine, and retried with backoff;
  • pooled browser profiles are wiped on every use, and each agent gets a distinct fingerprint;
  • an orphan reaper cleans up leftover browsers;
  • every browser DevTools command has a time limit, so no call can hang forever.
• New tools:
  • compare: identical or similar-page A/B diff across visual, DOM, text, network, schema, SEO,
    links, styles, a11y, console, headers and perf. It supports selectors, masks, normalization,
    noise calibration, parallel device/engine tracks and an HTML report.
  • schema: structured-data extraction, validation and diff.
  • cms: platform fingerprinting.
  • doctor: a read-only health check.
  • procs: a browser process census with orphan cleanup.
  • calls and cancel: in-flight call tracking.
  • config: every automatic setting can be tuned by hand. doctor, procs and config work
    without a license.
• smoke:
  • INFRA lines for launch failures, per-unit retry, a circuit breaker and --profile;
  • its existing PASS/FAIL and summary lines are unchanged.
• Optional priority controls: --priority high and --block images,media,fonts.
• One exit-code contract for every tool:
  • 0 ok
  • 2 usage error
  • 3 checked failure
  • 4 license required
  • 5 infrastructure failure (safe to retry)
  • 6 timed out or cancelled
• Security:
  • the license check fails closed;
  • signatures are verified with native Ed25519, after a startup self-test;
  • the compiled build ignores non-local endpoint overrides;
  • a license-bypass path was closed.
• Behaviour changes to note:
  • fetch, screenshot and dump default to a freshly wiped pooled profile. Use --profile NAME
    for a saved login.
  • An unreachable page is now a real failure (exit 3) instead of a false success.

Install

1. Run ChromeWalk-Setup.exe (per-user, no admin needed). On the ChromeWalk license page, paste
   your key or choose Email me a license key.
2. Run chromewalk connect to register ChromeWalk in the AI apps on this machine.
3. Claude Desktop: open the .mcpb instead, and paste your key into the extension's settings.
4. Run chromewalk doctor to check that everything is healthy.

If you're on 9.0: upgrade to 10.0. It includes a license-check hardening fix, and 9.0 will not
receive a separate update.

ChromeWalk 9.0.0

ChromeWalk 9.0.0

Real, local browser automation for your AI agents - exposed as a local MCP server
(chromewalk mcp, 18 tools) and a CLI. Runs entirely on your machine.

A license is required. Sign up at https://api.chromewalk.com/register.html (your key arrives by
email), or run chromewalk register you@example.com after installing, then
chromewalk activate <key>.

Downloads

File	For	SHA-256
ChromeWalk-Setup.exe	Windows 10/11 x64 - installs chromewalk for every AI app (Claude Code, Codex, Cursor, VS Code, Gemini CLI, ...)	9850d532c753e919bce1c4df851a6e16851642a4394fb0840528f5ac56e5658d
chromewalk-9.0.0-win32-x64.mcpb	Claude Desktop extension (Windows x64) - double-click to install	c217800b3baa907b6611eb7ff446fa15dc6feeb62e974b851172bf8cdeeb32f5

SHA256SUMS.txt lists both. macOS packages are in testing and not published yet.

The Windows installer is not code-signed yet (certificate pending), so SmartScreen may show
"Windows protected your PC" - choose More info -> Run anyway, or verify the SHA-256 first.

Install

1. Run ChromeWalk-Setup.exe (per-user, no admin). On the ChromeWalk license page, paste your
   key, or choose Email me a license key.
2. Run chromewalk connect to register ChromeWalk in the AI apps on this machine
   (or follow the per-app steps in the README).
3. Claude Desktop: open the .mcpb instead and paste your key into the extension's settings.

What's new in 9.0

• Compiled, self-contained program (no Python needed), ~0.1 s startup; Playwright engines
  (Chromium / WebKit / Firefox) installed by chromewalk setup.
• One launch command for every client: chromewalk mcp.
• In-app sign-up: chromewalk register <email>; installer license page.
• Email-verified licensing with a 2-device limit, new-device alerts, self-service device reset
  and key rotation.
• Works inside the Microsoft Store edition of Claude Desktop (browsers launch from their real location).
• Extras: chromewalk launch claude|chatgpt, chromewalk browsers, chromewalk add-extension.