# Read or extract data from a page that needs a real browser

> Read or extract data from a page that needs a real browser: 12 companies fully do it, 6 with a limit, 68 carry the tag. Every listing in Browser automation with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/browser.read-dynamic-page/
- Markdown: https://www.anchorterminal.com/niches/browser.read-dynamic-page/index.md (~1,950 tokens)
- Slim: https://www.anchorterminal.com/niches/browser.read-dynamic-page/index.min.md (~380 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/browser.read-dynamic-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

Four or more companies' docs show a full fit.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 12 | 6 | 68 | 18 | 12 |

## What counts as done

Structured fields or clean text returned for the page in one session, with the method reusable on the next page without another model call.

Measured, for our own tests: the result is correct against a manual read and returned in under a minute.

How agents ask for it: get the prices off this page; pull the listings from these search results; what does this dashboard say; scrape the Google Maps results for X

## Does the job

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

- **agent-browser (Vercel Inc.)**: Real Chrome renders JavaScript; get text with --max-output and snapshot -i return clean reads, and saved flows (experimental) replay a worked-out task's requests with parameters and no model. acc-47 cut cost by batching into client scripts.
- **Anchor Browser (Anchor Forge Inc.)**: perform-web-task takes a JSON schema and returns structured output from a rendered page, and the action index runs cached site actions as deterministic code with no LLM steps, reusable by name on the next run.
- **Browserbase (Browserbase, Inc.)**: Stagehand extract returns schema-shaped data from a cloud browser; acc-20 found extract never failed at about $0.001 for three atomic steps. Optimize turns a finished run into a Stagehand script for reuse, and acc-38 asked for the extraction plan to avoid a model call per page.
- **Browserless (browserless.io, Inc.)**: browserless_smartscraper picks fetch, proxy, headless browser or captcha solving and returns markdown or json in one call, and a BrowserQL query or browserless_function is deterministic code that reruns without a model.
- **BxPlaywright (Ortus Solutions, Corp.)**: text(), html(), snapshot() and evaluate() read a rendered page from BoxLang, and the script reruns on the next page without a model.
- **Chrome DevTools MCP (Google LLC)**: Real Chrome renders the page, take_snapshot and evaluate_script write large results to filePath, and evaluate_script's sourcePath or the chrome-devtools CLI reruns a worked-out read without a model.
- **Hyperbrowser (S2 Labs Inc.)**: web_fetch returns markdown or HTML of a rendered page and extract_json_from_webpage takes a JSON schema; Fetch is a fixed-price call that reruns without a model, and sessions take useStealth.
- **Kernel (Kernel Technologies, Inc.)**: Playwright over CDP against a browser with anti-detection defaults renders the page, and the same Playwright script reruns without a model; the MCP server executes Playwright code directly.
- **Notte (Notte Labs Inc.)**: browser_scrape returns markdown or structured data, observe lists actions with selectors, and session.execute runs typed actions with no model, with stealth, proxies and captcha solving on by default.
- **Playwright MCP (Microsoft Corporation)**: A real browser sees the rendered page (acc-02: sites aren't lying to Playwright), browser_snapshot and browser_evaluate write to filename, and a worked-out read reruns as Playwright code through browser_run_code_unsafe or the Playwright CLI; acc-05 saved repeated tasks as scripts.
- **Puppeteer (Google LLC)**: Real Chrome or Firefox renders the page, locators wait for elements, and the script reruns on the next page with no model call.
- **Selenium (Software Freedom Conservancy)**: A real browser renders dynamic pages and the script reruns without a model; acc-60 saved each dynamic Amazon page to an HTML file for Claude to read.
- **Steel (Nen Labs, Inc.)**: POST /v1/scrape returns a rendered page as Markdown with no session to manage, and a session over CDP runs the same Playwright script on the next page without a model.

## 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.

- **Airtop (Switchboard Visual Technologies, Inc. (dba Airtop))**: page-query takes an outputSchema and returns structured fields from a rendered page, so the page never enters the caller's context. Every call is an AI call priced in unpublished credits, and reuse on the next page without a model needs your own Playwright script.
- **Browser Use (Browser Use, Inc.)**: output_schema returns parsed JSON from a stealth cloud browser, but every run is a model run; a saved workspace script is still executed by an agent. acc-23 and acc-26 report success claimed with wrong data and high token cost.
- **Lightpanda (Selecy SAS)**: markdown, extract with a CSS-selector schema and structuredData read JavaScript pages, and a PandaScript saved from the session reruns with lightpanda run and no LLM. acc-64 measured 10 of 17 Playwright tests passing against Chrome's 17.
- **Pomerado (Pomerado AI, Inc.)**: A built read tool returns typed results in seconds and read-only runs can go in parallel. The first build explores the site and writes the tool, which is slow and counts against 10 new integrations a month on Free.
- **Skyvern (Ikonomos Inc.)**: extract returns data shaped to a JSON schema, and Run With Code replays the generated script without AI once cached, with AI fallback on failure. acc-52 found it did not work for scraping one site, and the first run pays 30 credits an action.
- **TinyFish (Tiny Fish, Inc.)**: output_schema shapes the result and the stealth profile helps on protected sites, but every run is model steps with no saved replay, and the docs say a COMPLETED run with an empty or wrong result is the common anti-bot failure.

- not this job: CurlShot (MEDIAPRONET SRL): Output is an image, PDF or video; no text, markdown or structured fields.
- not this job: Functionize MCP server (Functionize, Inc.): A test platform: values can be captured into variables and verified, but nothing returns extracted page data to the caller as the product's job.
- not this job: Kane CLI (TestMu AI (formerly LambdaTest)): A test runner that returns pass or fail with evidence; captured values feed checkpoints, not an extraction result for the caller.

## What stops agents finishing

- a bot check the tool cannot pass (see browser.pass-bot-check)
- an agent that reports success with wrong data and no step log (acc-26)
- a page whose snapshot alone fills the context window (acc-19, acc-20)
- a lighter engine that opens the page but cannot click the link (acc-64)

## Around it

Before: search.fetch-url, web-scrapers.page-to-markdown. After: web-scrapers.price-and-change-monitoring, document-extraction.table-from-text.

## Where this comes from

Weight 18 of 100 in Browser automation: 12 of 61 accounts; three of them had first tried a plain scraper or a scraping tool and moved to a browser (acc-02, acc-06, acc-23, acc-25, acc-26, acc-30, acc-36, acc-38, acc-47, acc-56, acc-60, acc-64)

As data: https://www.anchorterminal.com/api/v1/use-cases/browser.read-dynamic-page.json, or the MCP tool get_use_case with id browser.read-dynamic-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
