ChromeWalk

Cross-engine gate cw xengine

3 engines Shared browser checks

BL

Chromium Functional + render passed

GK

Firefox Functional + render passed

WK

WebKit Functional + render passed

built-in detectors × 3 engines · divergence report · JUnit CI gate

Detailed browser capabilities resolve into compact tool cards linked to a full local artifact.

ChromeWalk · flow coverage Scripted checks and safe exploration lead to a coverage report, with destructive controls isolated. Conceptual illustration.

Render JavaScript-heavy pages, investigate browser failures, and collect structured data with one local toolkit.

ChromeWalk uses its own browsers and profiles. Capture screenshots, network responses, and page data without mixing automation with your everyday browsing.

35 AI-callable tools across capture, debugging, auditing, commerce, recovery, and operations

From one request to an entire evidence pipeline.

ChromeWalk does more than open a page. It routes structured work from an MCP client into isolated browsers, governs machine capacity, captures the page and its surrounding systems, and sends reviewable evidence back to the person or agent who asked.

01 AI client 02 Local MCP 03 Resource governor 04 Browser fleet 05 Evidence returns

REVIEWABLE ARTIFACTS

Keep the page and the evidence together.

A ChromeWalk design audit preserves the rendered page, its screenshot and structured snapshot records. These artifacts let another person inspect the source behind the findings.

Workflow diagram: save the page, screenshot, and structured snapshot together for later review.

• Connect an agent to local tools →
• Capture, inspect and manage work →
• Compare independent browser engines →
• Choose an evidence workflow →

Connect your agent. Capture the page. Check the result.

Choose from 35 MCP tools, connect one of 16 supported clients, and run checks across Chromium, Firefox, and WebKit. Each part helps you turn a browser task into useful evidence.

35 · MCP TOOLS

Thirty-five callable capabilities.

These are the public agent tools for capture, debugging, audits, comparison, commerce, recovery, and operations.

Count this as

Ask your agent to research, capture, debug, compare, or audit a page.

16 · CLIENT TARGETS

Sixteen auto-configured MCP clients.

cw connect supports Claude Desktop, Claude Code, Codex, Cursor, Windsurf, Cline, Gemini CLI, VS Code, Zed, Continue, GitHub Copilot CLI, Kiro, Amazon Q Developer, Kilo Code, Goose, and Visual Studio.

Count this as

Connect your preferred client with its step-by-step setup guide.

3 · BROWSER ENGINES

Three real cross-engine runtimes.

Chromium, Firefox, and WebKit run the shared functional and structural verification contract.

Count this as

Find behavior that differs between Chromium, Firefox, and WebKit.

2 · CHAT-SITE WINDOWS

Two separate chat-site windows.

Open claude.ai or chatgpt.com in a separate ChromeWalk window that your agent can read. This is not a ChatGPT integration.

Count this as

Keep assistant browsing in dedicated profiles for repeatable work.

12 · QA DETECTORS

Twelve distinct failure rules.

They look for console and page errors, rejected promises, "undefined" text, broken click handlers and inputs, content wider than the screen (at load and before the page settles), elements escaping the top-right corner, modal overlap, and mobile overlays on desktop.

Count this as

Catch common page failures before users encounter them.

36 · ENGINE CHECKS

Twelve detectors across three engines.

The cross-engine self-test runs the same 12 rules in Chromium, Firefox, and WebKit: Cross-engine checks.

Count this as

Check that each detector can identify its expected failure in every engine.

Sixteen supported client targets. Sixteen detailed guides.

ChromeWalk speaks local stdio MCP, so 16 supported AI apps (15 on a Mac: Visual Studio is Windows only) can discover the same 35 MCP tools. The highlighted cards below are a quick starting point; the connector directory documents all 16 clients.

From stubborn page to undeniable result.

Choose the mission. ChromeWalk brings the command, the evidence, and the guardrails—then gets the fleet moving.

01 / 06

Capture the entire page universe.

Pull in every endpoint, stylesheet, script, font, image, and response body—then add viewport and full-page screenshots. Large bodies stream in chunks and blobs are SHA-sharded, so enormous captures stay controlled.

• Real Chromium takes on JavaScript challenges
• Parallel URLs get unique fingerprints
• Evidence lands the instant each resource finishes

› cw dump https://site.tld --lean --full-page --json

DEPLOYMENTcw-8f32ENGINE      ENGINE Chromium · headless-max
IDENTITY    win-nvidia · isolated

✓
 164 resources cataloged
✓
 159 bodies captured
✓
 2 screenshots written
✓
 manifest.jsonl complete

Run complete                          2.8s

Spend less time repeating browser checks.

⌘

Orchestration

Go massively parallel. Leave collisions behind.

A live memory governor sizes the worker pool while cross-process fingerprint leases, atomic profile claims, and unique output paths keep every agent in its own lane.

◈

Privacy

Keep automation separate from everyday browsing

ChromeWalk runs its own browser binary and profiles. Browser sync, sign-in, and browser background services are disabled by default. ChromeWalk usage analytics are required by its license; see the Privacy notice.

Personal Chrome Untouched

◎

Sessions

Warm once. Unleash the fleet.

Sign in yourself, save a named session, and send cookies or origin-guarded local storage across fetch, dump, drive, and smoke runs.

--session account

▦

Commerce intelligence

Turn entire catalogs into structured intelligence.

Harvest rendered product data across SPA storefronts into normalized JSONL or CSV: identity, price, availability, images, specifications, variants, metafields, and Year/Make/Model fitment. Add screenshots and store facets in the same visit.

Full catalog products in a verified live full-catalog run

Products Screenshots Facets CSV

◇

Verification

Catch common page failures before release.

Run checks for page errors, broken handlers, overflow and overlapping elements, and add your own required-element checks. Use the detector self-test to confirm the checks can identify their expected failures.

console errorspage errorsunhandled promise rejections"undefined" textundefined click handlersundefined inputshorizontal overflowelement wider than the screentop-right escapeoverflow before the page finishes loadingmodal overlapmobile overlay on desktop

More signal from every browser minute.

ChromeWalk is useful when a research script needs rendered truth, a debugger needs reproducible evidence, or a catalog job needs structured output without repeated visits.

01 · PROGRAMMATIC RESEARCH

Search past the static HTML.

Load the page as a browser sees it, wait for rendered content, then return clean text, status, validation, and JSON. Researchers can cover JavaScript-heavy sources that a plain request cannot fully represent.

Benefit

Fewer empty responses, fewer manual browser checks, and structured results that move directly into an analysis pipeline.

02 · PARALLEL DISCOVERY

Turn a URL list into a governed fleet.

Distribute fetch, dump, screenshot, or harvest work across isolated agents. Unique fingerprints, profile claims, output directories, and live resource sizing keep parallel runs from colliding.

Benefit

Cover more sources at once without making machine stability or result ownership somebody else’s problem.

03 · FORENSIC DEBUGGING

Capture the evidence, not just the symptom.

Collect console failures, rejected promises, HAR activity, Core Web Vitals, accessibility data, raw response bodies, screenshots, and layout geometry around the same application state.

Benefit

Replace “it broke somewhere” with a durable evidence bundle a developer or agent can inspect and reproduce.

04 · CROSS-ENGINE QA

Catch the bugs Chrome never shows you.

Run the same functional suite through Blink, Gecko, and WebKit, then report which engine diverged and which check failed. Optional render gating compares landmark geometry instead of noisy antialiasing.

Benefit

Find engine-specific regressions before release and feed standard JUnit results into CI.

05 · COMMERCE INTELLIGENCE

Convert rendered catalogs into usable data.

Normalize products, variants, prices, availability, images, specifications, metafields, breadcrumbs, and vehicle fitment. Pull store-level facet axes and screenshots in the same crawl.

Benefit

Create research-ready JSONL or CSV with fewer page visits and a coverage report that shows what was actually captured.

06 · AUTHENTICATED WORK

Reuse access without scripting secrets.

Sign in yourself through a visible browser, save the session, and reuse cookies or origin-guarded local storage across authorized research and QA runs.

Benefit

Repeat role-based workflows across agents without typing credentials into automation or touching another browser’s stored data.

Install ChromeWalk Windows · macOS (Apple Silicon) · Claude Desktop local-first

1. 01 Download Download ChromeWalk, or run the one-line installer from the download page. download
2. 02 Get your key Get your key by email, then activate it from your AI assistant or with chromewalk activate KEY (add --remember on Windows). email
3. 03 Connect Run chromewalk connect, then chromewalk mcp-selftest. local

What works offline

Local apps, local targets, local evidence.

App registration, the control panel, MCP discovery, connector self-tests, saved artifacts, JSON queries, and browser work against local environments can stay on the machine after provisioning.

Portable by design

Move a provisioned browser operation wherever the work goes.

Keep the runtime, profiles, sessions, logs, evidence, and MCP configuration together. Teams can reproduce the same workflow across controlled workstations without rebuilding the operating model.

The marketing advantage

Keep your browser workflow under your control.

Run browser jobs on your workstation, keep the resulting files close to your project, and reuse the same workflow from an agent, a script, or CI.

Build it. Launch it. Watch it fly.

Pick the mission and ChromeWalk assembles a safe starting command. One-time setup, then the whole browser fleet is at your fingertips.

Windows + macOSSelf-contained appStandalone browser fleet

cw fetch https://example.com --lean --json

Need installation? Run the native Windows or macOS setup, connect supported clients with cw connect, then verify locally with cw mcp --selftest.

What is new in ChromeWalk

Content and SEO audits, hidden-break reports, respectful pacing and lean AI calls.

See the latest additions

Start in minutes. Master the entire machine.

The source documentation is the product record. These guides rocket you straight to the right workflow without burying the answer under a mountain of flags.

GET STARTED · 4 MIN

Quickstart for agents

Install the standalone runtime, run the first smoke checks, and learn the three exit codes.

GET STARTED · 6 MIN

Install & connect

Use the native Windows or macOS setup, then connect supported MCP clients.

GET STARTED · 5 MIN

Offline-capable deployment

Separate one-time provisioning from offline app registration, local MCP verification, and operation.

RUN & CAPTURE · 8 MIN

Fetch vs. full dump

Choose the smallest capture mode that preserves the evidence your task needs.

RUN & CAPTURE · 7 MIN

Multi-agent smoke suites

Fan pages across viewports with isolated profiles and self-tested detectors.

RUN & CAPTURE · 5 MIN

Sessions & authenticated runs

Warm a login yourself, save it by name, and reuse it without credential scripting.

COMMERCE · 10 MIN

Harvest product catalogs

Extract normalized products, screenshots, facet axes, and CSV in one browser pass.

VERIFY · 9 MIN

Cross-engine CI gates

Run the proven detector library across Blink, Gecko, and WebKit with structural render checks.

BUILD & CONNECT · 9 MIN

Connect through MCP

Expose all 35 ChromeWalk tools to an AI client over a zero-dependency stdio server.

BUILD & CONNECT · 7 MIN

AI coding-agent connectors

Configure the same local ChromeWalk server in 16 supported AI apps; all sixteen have dedicated public guides.

BUILD & CONNECT · REFERENCE

CLI command map

Move from cw fetch to render, drive, inspect, API, PWA, and logs.

REFERENCE · LIVING DOCS

Human & AI manuals

Use code-generated references that stay synchronized with the MCP registry and every tool’s help.

SAFETY · READ FIRST

Boundaries & guardrails

Understand human verification, TLS limits, privacy defaults, and legitimate-use boundaries.

Cataloged for humans. Legible to agents.

ChromeWalk now exposes a public-facing inventory of every MCP surface, crawler-friendly product context, and a documented route toward official AI directories.

Website discovery Ready

Local stdio MCP 35 tools

Feature intelligence 35 MCP tools

Script-grounded reference Live schemas

One durable map of the ChromeWalk universe

AI systems can discover what ChromeWalk does, select the right command or MCP tool, understand the returned artifact, and move directly from intent to a grounded execution plan.

Capture fetch · screenshot · render · dump

Inspect webinspect · pwa · jsonquery · api

Validate smoke · xengine · flow

Commerce harvest · facets

Operate drive · notify · logs

Discover resources · sites_list

01 Discover

Browse every capability by outcome, benefit, and real usage scenario.

02 Decide

Compare commands, inputs, outputs, and workflows before launching a browser.

03 Execute

Run through the CLI or give an AI agent the same structured MCP tool.

04 Verify

Return JSON, screenshots, manifests, reports, and evidence ready for the next decision.

ChromeWalk’s public catalog, Detailed feature pages, command examples, MCP schemas, and AI discovery files describe the same shipped ChromeWalk surface.

Answers for humans. Context for agents.

Each answer is written for scanning, indexed as structured FAQ data, and mirrored in the dedicated AI discovery files.

llms.txt llms-full.txt XML sitemap

How does ChromeWalk improve programmatic research?+

It renders JavaScript-heavy pages in a real isolated browser, supports warmed sessions when access is authorized, and emits pure JSON on stdout. Batch and multi-agent modes let a research process cover many sources while results stream to disk instead of accumulating in memory.

What makes it useful for debugging?+

ChromeWalk joins the artifacts developers usually collect separately: console and page errors, unhandled rejections, HAR, Web Vitals, accessibility data, screenshots, raw response bodies, functional assertions, and layout geometry. The result is a reproducible evidence trail rather than a single screenshot.

What does the cross-engine gate actually test?+

cw xengine runs the same 12-detector functional library in real Chromium, Firefox, and WebKit engines. It reports per-engine outcomes and divergences. With --render, it also compares structural landmark geometry and writes a JUnit report for CI.

How does cross-engine testing sharpen release confidence?+

One suite runs through Chromium, Firefox, and WebKit with the same detectors and assertions. ChromeWalk highlights engine-specific failures, compares structural geometry, and produces JUnit output that can gate a release.

Can it search and extract whole commerce catalogs?+

Yes. cw harvest can seed from a sitemap or URL list, render SPA product pages, and normalize product identity, prices, availability, images, specifications, variants, metafields, and Year/Make/Model fitment. Screenshots, facets, and CSV are combinable levers in the same pass.

How does ChromeWalk avoid parallel-agent collisions?+

It combines a live RAM governor with cross-process fingerprint leases, atomic persistent-profile claims, isolated fallback profiles, and unique output directories. The design favors a slower safe run over an unstable overcommit.

How does ChromeWalk keep multi-agent browser work organized?+

Every run can use isolated profiles, leased identities, unique output directories, and RAM-aware concurrency. The result is clean ownership of sessions and artifacts even when several agents are working at once.

How do warmed sessions accelerate authenticated work?+

Sign in once through a visible browser, save the authorized state by name, and reuse it across fetch, dump, drive, smoke, and harvest workflows. Agents move straight to the work while the session remains under operator control.

How do AI agents discover the complete tool surface?+

Use llms.txt for the compact route map, llms-full.txt for the capability and safety reference, and sitemap.xml for machine-readable discovery. Inside the product repository, the ChromeWalk documentation is generated from the live MCP registry and tool help.

How can I integrate ChromeWalk into an AI client or CI system?+

AI clients can connect to the local stdio MCP server and call 35 tools. CI systems can run the CLI, consume JSON and exit codes, publish JUnit gates, and retain self-contained HTML audit or comparison reports.

Which AI coding agents can connect to ChromeWalk?+

Any client that launches a local stdio MCP server can use the connector. Dedicated guides cover all 16 supported clients.

How quickly can ChromeWalk become part of a workstation?+

Install per user without administrator rights, get a license key by email with cw register, activate it, connect preferred coding agents, and check the AI connection with cw mcp-selftest, which starts ChromeWalk's server, lists its 35 tools and runs one.

Is the ChromeWalk license available by email?+

Yes. Get a license key by email through the signup page or cw register you@example.com, then activate the workstation with cw activate <key>. A license is required for browser operations, while key diagnostics and recovery commands remain available so a license or installation problem can still be fixed.

Is ChromeWalk available now?+

Yes. ChromeWalk is available for Windows as an installer and as a Claude Desktop extension, and for Apple Silicon Macs as a package. The Windows app updates itself; on a Mac and in Claude Desktop you are told when an update is ready.

Where can teams explore every capability in depth?+

The feature catalog links to an individual public page for every MCP tool with benefits, usage cases, examples, outputs, and current arguments. The installed cw help reference carries the exhaustive command detail without exposing private service implementation on the public site.

What stays on my computer, and what does ChromeWalk send?+

The pages ChromeWalk opens, its screenshots and files, your browser profiles and logins, and your Nimble and proxy passwords stay on your computer. ChromeWalk sends licence checks, update checks and usage events (which tool ran, whether it worked, how long it took, the option names used and the web address with logins, query strings and fragments removed). It never sends page content, files, file paths, option values, passwords or your licence key. Run cw telemetry status to see exactly what is sent.

Test the flows. Start in chat. Reach more browsers.

Flows, not just pages.

Each scripted step can carry an expect: a selector, absent element, text within a region, or URL change. New console errors and uncaught exceptions are attributed to the step; same-origin 5xx and network failures fail it, while 4xx responses are warnings. A missing step target now fails the page. Add --explore 25 --coverage to smoke or xengine. Explore clicks visible controls that pass its safety filters, then restores the page. The default time cap is 20 seconds per page. Destructive-looking controls, form submits, downloads, new-tab links and offsite links are skipped. It does not fill forms or complete multi-page journeys. Use staging or a test account; an explicit allow regex can override label filtering.

Step checks and coverage ↗

Zero-friction start.

An MCP assistant can call chromewalk_register with your email and chromewalk_activate with your key before activation. The key is emailed, never returned by registration, and activation never echoes it. The optional Claude Desktop License Key field can remain empty for in-chat onboarding. Pasting a key into a chat exposes it in that conversation; use the extension setting or terminal if you prefer.

Activate in your assistant ↗

More browsers, more AI tools.

One local stdio server exposes 35 MCP tools; the CLI documents 59 commands. cw connect supports 16 targets, including the six new targets GitHub Copilot CLI, Kiro, Amazon Q Developer, Kilo Code, Goose and Visual Studio. Configuration is backed up before the first change, unrelated entries are preserved, and your key is never written into client configuration. Roo Code and JetBrains need manual settings; cloud chat apps cannot reach a local program without a separate supported connection. Choose --browser edge, brave, chrome or a Chromium executable path for smoke, fetch, dump, screenshot, compare, design and audit. Chrome for Testing remains the default. Edge and Brave are Chromium browser choices, not additional engine families. Every run uses a ChromeWalk-owned profile; your personal history, logins and extensions stay separate. Brave Shields can change requests and page behavior. Missing installed browsers produce an engine_missing error; ChromeWalk does not install them.

Choose your connection ↗

More insight. Leaner calls.

Reuse the text your browser already captured. Content and SEO metrics use no AI tokens. Start with core arguments; load advanced help only when the job needs it.

Content Analysis: Words as Well as Pixels

Measure readability, compare prose and find duplicate pages using text already captured by ChromeWalk. Deterministic local metrics use no AI tokens.

Captures That Explain Themselves

Give screenshots and saved files meaningful names, and trace each capture to its page and step with a local navigation log.

SEO Audits of Real Browser Pages

Audit rendered titles, descriptions, links and structured data. Get ranked SEO fixes and check the pages you choose from your AI assistant or CLI.

Catch What Breaks Quietly

Collect failed API calls, broken scripts and images, policy blocks and unhandled errors during browser walks, with findings tied to the affected page.

Polite Per-Site Pacing

Respect rate limits and Retry-After signals with shared per-host pacing. Healthy sites keep moving while a busy host gets time to recover.

Keep Long Browser Walks Useful

Trigger lazy page content, keep active runs alive and detect expired sessions. ChromeWalk records one bounded recovery instead of repeating a login wall.

Works with Nimble: Optional Regional Routing

Opt in to regional routing with your own Nimble account. It is off by default, and your local browser does the work through the optional network.

Know Exactly What Runs in Each Engine

Check the generated capability matrix for Chromium, Firefox and WebKit. See supported tools and options, with clear reasons for engine-specific gaps.

Full Browser Tools, Lean AI Context

Use concise MCP tool listings, on-demand help and compact replies linked to full local results. ChromeWalk starts no workers or background checks while idle.

Load the detail when you need it.

ChromeWalk exposes 35 MCP tools through concise descriptions and core arguments. Brief help gives the essentials; full help opens the complete reference without launching a browser. Large results return a summary and a path to the full local file. The idle server starts no workers, timers or background checks.

Content metrics reuse captured text, and prose comparisons over local files need no browser. Optional features cost no work when you leave them off.

See lean calls and on-demand help