# Debug or profile a live web app from its console, network and traces

> Debug or profile a live web app from its console, network and traces: 2 companies fully do it, 8 with a limit, 30 carry the tag. Every listing in Browser automation with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/browser.debug-live-page/
- Markdown: https://www.anchorterminal.com/niches/browser.debug-live-page/index.md (~1,800 tokens)
- Slim: https://www.anchorterminal.com/niches/browser.debug-live-page/index.min.md (~330 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/browser.debug-live-page/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

Two or three companies' docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 2 | 8 | 30 | 10 | 7 |

## What counts as done

The agent names the failing request, console error or long task with its source location, from one trace or network capture, and the data lands in a file rather than the context when it is large.

How agents ask for it: why is this page slow; what is failing in the console; find the memory leak; which request returns the 500

## Does the job

The docs show it meets the test above, and nothing we read says otherwise.

- **agent-browser (Vercel Inc.)**: console and console --json read page errors, network har stop and trace stop write files, profiler gives curated performance categories and --auto-connect attaches to the person's running Chrome. acc-41 runs it beside chrome-devtools-mcp.
- **Chrome DevTools MCP (Google LLC)**: performance_start_trace writes trace.json.gz and performance_analyze_insight names LCP and other insights, list_network_requests pages results and get_network_request saves bodies to files, console messages carry source-mapped stacks, and heap tools paginate (acc-43 closed completed). acc-37 used it to profile a slow React page.

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Anchor Browser (Anchor Forge Inc.)**: The anchor_* MCP tools expose console and network logs over a live session, but no performance trace, profiler or heap tools, and the browser is remote so the agent's own app needs a tunnel. Enough to name a failing request, not a long task.
- **Browserbase (Browserbase, Inc.)**: The browser-trace skill records a CDP trace, screenshots and per-page events to .o11y/<run-id>/ and lists failed requests and console errors. It creates a keep-alive session, which is paid-plan only, and a remote browser needs the app reachable from the cloud.
- **Browserless (browserless.io, Inc.)**: browserless_performance runs a Lighthouse audit and session replay records console logs and network requests for the dashboard. No trace or heap tools, and the cloud browser cannot see a localhost app.
- **BxPlaywright (Ortus Solutions, Corp.)**: A Network and APIs page and a trace viewer (show-trace trace.zip) exist, but no console, performance or heap tooling was read, and the output is a Playwright trace rather than a named failing request.
- **Kane CLI (TestMu AI (formerly LambdaTest))**: The evidence pack holds console NDJSON and a HAR per step on disk, with failure records pointing into them, and DevTools checkpoints cover console, network and performance. No trace or profiler for finding a long task.
- **Playwright MCP (Microsoft Corporation)**: browser_console_messages and browser_network_requests filter by level or regexp and save to a file, browser_network_request returns one request's details, and --save-trace writes a Playwright trace. No performance insights or heap tools; acc-37 doubted it could profile a slow page.
- **Puppeteer (Google LLC)**: page.on request and response events log network, console events are available and headless false plus slowMo help watch a page, but traces and heap data need raw CDP code; the docs send agents to chrome-devtools-mcp for that.
- **Selenium (Software Freedom Conservancy)**: WebDriver BiDi streams console entries, JavaScript exceptions and network events with interception, but there is no performance trace or heap tool, and the output goes wherever the script writes it.
- **Steel (Nen Labs, Inc.)**: Each recorded session has Console Logs and Network tabs and an Agent Traces timeline with JSON and ZIP exports, but no performance trace or heap tools, and the cloud browser cannot see a localhost app.

- not this job: Airtop (Switchboard Visual Technologies, Inc. (dba Airtop)): The debugging guide describes a person opening DevTools inside the Live View by typing debugger in its URL bar. No console, network or trace endpoint for the agent, and the browser runs in Airtop's cloud.
- not this job: Browser Use (Browser Use, Inc.): Network capture means attaching your own CDP observer to a cloud browser; no console, trace or profiler tools, and the cloud browser cannot see the agent's localhost app.
- not this job: CurlShot (MEDIAPRONET SRL): No console, network or trace output; the service renders and captures only.
- not this job: Functionize MCP server (Functionize, Inc.): Failure analysis uses screenshots, steps, timing and network inside Studio, but no console, trace or request data is exposed to the coding agent as tools.
- not this job: Hyperbrowser (S2 Labs Inc.): No console, network or trace tools; the cloud browser is reached over CDP and the agent's own app needs a tunnel.
- not this job: Kernel (Kernel Technologies, Inc.): Infrastructure with replays and a REPL, no console, network or trace tools, and the browser sits in Kernel's cloud away from the agent's app.
- not this job: Lightpanda (Selecy SAS): consoleLogs is the only debugging tool; no network list, trace or profiler, and no rendering to profile.
- not this job: Notte (Notte Labs Inc.): No console, network or trace tools; the cloud browser is reached by CDP or typed actions.
- not this job: Pomerado (Pomerado AI, Inc.): No console, network or trace output; the agent never sees the browser.
- not this job: Skyvern (Ikonomos Inc.): The MCP mentions network and console inspection tools, but there is no trace or profiler and the product is a task agent, not a debugger for the agent's own app.
- not this job: TinyFish (Tiny Fish, Inc.): No console, network or trace output; the Browser API hands over a CDP URL and nothing more.

## What stops agents finishing

- a heap snapshot too large for an agent to read (acc-43)
- a permission dialog on every connection to the person's Chrome (acc-42)
- the browser stealing focus on every command while the person works (acc-46)

## Around it

Before: agent-harnesses.pair-in-terminal, browser.check-web-app. After: agent-harnesses.spec-to-build-loop.

## Where this comes from

Weight 10 of 100 in Browser automation: 7 of 61 accounts, all chrome-devtools-mcp or Puppeteer users (acc-37, acc-41, acc-42, acc-43, acc-46, acc-58, acc-63)

As data: https://www.anchorterminal.com/api/v1/use-cases/browser.debug-live-page.json, or the MCP tool get_use_case with id browser.debug-live-page. Draft, read 2026-11-12 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
