{"slug":"cloudflare-os-browser","title":"Browser Rendering is an evidence adapter, not a better fetch()","body":"# Browser Rendering is an evidence adapter, not a better `fetch()`\n\nA plain HTTP client retrieves bytes. Cloudflare Browser Run can execute the page, wait for its state to settle, and return a representation chosen for the next operation: rendered HTML, Markdown, selected elements, links, a screenshot, a PDF, an accessibility tree, structured JSON, or an asynchronous crawl.\n\nThat distinction is the whole chapter. A browser belongs in this system only where the evidence depends on browser execution or a browser-specific representation. It should not become the default transport. Making it the default spends more time and money, enlarges the security boundary, and can still return a convincing but incomplete page.\n\nThe capability catalogue therefore does not contain one vague `BROWSER` tool. It contains explicit contracts: what representation is requested, what completion condition is required, what authority may leave the system, and what receipt must come back. The browser is the eyes. The canonical catalogue decides when those eyes may open and what counts as seeing.\n\n## Evidence status\n\n**Observed** marks first-party measurements or runtime receipts from the named environment.\n**Derived** marks arithmetic calculated from cited inputs. **Specified** marks vendor or standards\ndocumentation. **Implemented** and **deployed** name code and live-state evidence, respectively.\n**Reproduced** means the stated procedure was rerun. **Externally attested** marks operator reports;\nthose reports show that an experience occurred, not that it is universal.\n\n## The endpoint is a choice about evidence\n\nCloudflare exposes ten Quick Actions in the current documentation, including the beta crawl action. They overlap at the input—usually a URL—but not at the output. Choosing by convenience rather than by evidence type is how a screenshot gets mistaken for data, a Markdown conversion gets mistaken for the DOM, or a link inventory gets reconstructed expensively from a general browser session.\n\n| If the next operation needs | Quick Action | Returned evidence | Do not infer |\n| --- | --- | --- | --- |\n| Executed document markup | `/content` | rendered HTML | that every lazy region loaded |\n| Human-readable text and links | `/markdown` | converted Markdown | pixel layout or exact DOM fidelity |\n| Named fields from known selectors | `/scrape` | selector results | completeness outside those selectors |\n| Link discovery | `/links` | extracted links | that every destination is safe or relevant |\n| Visual state | `/screenshot` | raster image | semantic structure or hidden text |\n| Printable artifact | `/pdf` | PDF bytes | browser-screen layout |\n| Accessible semantic structure | `/accessibilityTree` | roles, names, states, children | that inaccessible controls do not exist |\n| Several representations together | `/snapshot` | two or more requested formats | that the formats agree automatically |\n| Schema-shaped extraction | `/json` | model-produced JSON | deterministic parsing or factual truth |\n| Multiple pages over time | `/crawl` | asynchronous crawl results | current unlimited throughput |\n\n`/snapshot` is especially useful for evidence work because one browser state can yield a visual surface and structural surfaces together. Cloudflare says the action defaults to HTML plus screenshot and can add Markdown and the accessibility tree. That is not just fewer requests. It reduces the chance that two captures were made from different page states. The receipt should still record each format separately and hash the bytes separately, because a screenshot and HTML prove different things.\n\nThe inverse rule matters too. If a stable endpoint already returns JSON, call it with ordinary HTTP. If static HTML contains the needed text, use ordinary HTTP. If all that is required is a status code or header, a browser weakens the measurement by adding navigation, rendering and conversion work that the question never asked for.\n\n## A browser can execute a page without proving the page is complete\n\nJavaScript execution is necessary for many modern pages, but it is not a completion oracle. Single-page applications often paint an initial shell, issue more requests, then reveal content after a selector appears. Cloudflare's Quick Action documentation repeatedly warns that the default result may be incomplete for SPAs and points to `waitForSelector` or navigation wait options.\n\nThat means every browser capability needs an explicit completion contract. “Open this URL” is not one.\n\n| Completion contract | Good for | Failure it prevents |\n| --- | --- | --- |\n| `waitUntil: \"domcontentloaded\"` | server-rendered page with small client enhancement | waiting for irrelevant long-lived connections |\n| `waitUntil: \"networkidle0\"` | bounded application that becomes quiet | capturing before dependent requests finish |\n| `waitForSelector: \"#results\"` | a known state transition | treating the application shell as the result |\n| fixed delay | almost nothing by itself | none; it only moves the race |\n| application assertion | login, checkout, dashboard state | proving the wrong authenticated or error state |\n\nA useful row therefore separates navigation from success:\n\n```json\n{\n  \"key\": \"BROWSER_MARKDOWN\",\n  \"what\": \"Return Markdown after the named page state exists.\",\n  \"args\": {\n    \"url\": \"https URL\",\n    \"wait_for_selector\": \"optional CSS selector\",\n    \"timeout_ms\": \"bounded integer\"\n  },\n  \"authority\": {\n    \"hosts\": [\"developers.cloudflare.com\"],\n    \"cookies\": false,\n    \"custom_headers\": []\n  },\n  \"receipt\": {\n    \"final_url\": true,\n    \"status\": true,\n    \"browser_ms\": true,\n    \"body_sha256\": true,\n    \"selector_observed\": true\n  }\n}\n```\n\nThe row is discoverable because its `what` names Markdown and page state. It is invokable because the arguments are concrete. It is auditable because the allowed hosts and credential channels are visible. It is replayable because the receipt records the final URL, completion observation and content hash. The same row can project to REST documentation, a model tool schema, a CLI command and an admin form without inventing four contracts.\n\n## The first-party receipt: 208 browser milliseconds, not a speed claim\n\nOn 26 July 2026 this build called the production `/browser-rendering/markdown` REST action against `https://example.com`. The token was read from the local credential store and was never copied into the artifact. The response was successful and contained the expected “Example Domain” heading.\n\n| Fresh measurement | Result |\n| --- | ---: |\n| Cloudflare response status | 200 |\n| Client-observed elapsed time | 1,418 ms |\n| `X-Browser-Ms-Used` | 207.706 ms |\n| API response bytes | 199 |\n| Returned Markdown characters | 167 |\n| Expected heading present | yes |\n\nThis is a receipt for one request from one client to one stable target. It is not a latency benchmark, an availability claim, or evidence that arbitrary protected sites will render. Client elapsed time includes network and API overhead. The browser-time header measures billable browser work for that Quick Action, not total wall time.\n\nThe reproduction is deliberately small:\n\n```bash\ncurl -X POST \\\n  \"https://api.cloudflare.com/client/v4/accounts/$ACCOUNT_ID/browser-rendering/markdown\" \\\n  -H \"Authorization: Bearer $BROWSER_RENDERING_TOKEN\" \\\n  -H \"Content-Type: application/json\" \\\n  --data '{\"url\":\"https://example.com\"}'\n```\n\nThe portable version should read the account identifier and token from environment or a secret store, never from a catalogue row, prompt, receipt or shell history. Record the response status, final representation hash and `X-Browser-Ms-Used`; discard the bearer token before ledgering.\n\n## The bill is browser time, and sessions add a second meter\n\nCloudflare distinguishes Quick Actions from Browser Sessions. Quick Actions are charged for browser hours. Direct sessions through Puppeteer, Playwright or CDP are charged for browser hours and, on paid plans above the included allowance, concurrent browsers.\n\nThe current published table gives Workers Free ten browser minutes per day. Workers Paid includes ten browser hours per month and then charges $0.09 for each additional browser hour. Browser Sessions include three concurrent browsers on Free; Paid includes ten averaged monthly and then lists $2 for each additional concurrent browser. The Quick Action response header reports browser milliseconds used, which is the useful per-invocation receipt field.\n\n| Cost or limit surface | Workers Free | Workers Paid default |\n| --- | ---: | ---: |\n| Browser time | 10 minutes/day | 10 hours/month, then $0.09/hour |\n| Quick Action rate | 1 request/10 seconds | 10 requests/second |\n| Session browsers | 3 concurrent | 120 concurrent limit |\n| Included session concurrency for pricing | 3 | 10 monthly-average daily peak |\n| New session instances | 1 every 20 seconds | 1/second |\n| Inactivity timeout | 60 seconds | 60 seconds |\n| Configurable inactivity timeout | up to 10 minutes | up to 10 minutes |\n\nThe 120-browser paid limit and the ten-browser paid price inclusion answer different questions. Conflating them makes a cost table wrong. So does multiplying the 208 ms receipt by the $0.09 rate and presenting the fraction of a cent as an invoice: Cloudflare aggregates daily seconds and rounds the monthly browser-hour total. The individual header supports attribution and anomaly detection; billing still follows the aggregate rules.\n\nDirect sessions need stricter lifecycle code:\n\n```js\nlet browser;\ntry {\n  browser = await puppeteer.launch(env.BROWSER);\n  const page = await browser.newPage();\n  await page.goto(target, { waitUntil: \"networkidle0\" });\n  return await page.content();\n} finally {\n  if (browser) await browser.close();\n}\n```\n\nCloudflare warns that a session left open continues consuming browser time until the inactivity timeout. An issue in `cloudflare/workers-sdk` also reported `browser.close()` hanging under local Vite and Wrangler development while production worked. That report is one historical local-development reproduction, not evidence that current production close calls hang. It is enough to justify a bounded close operation, a recorded close reason and a test of local and deployed paths separately.\n\n## “Cloud browser” does not mean “bypass”\n\nChanging the User-Agent does not turn Browser Run into an unidentifiable residential client. Cloudflare states that Browser Run requests are always identified as bot traffic and that a custom User-Agent does not bypass bot protection. A remote Chrome may execute client JavaScript that plain fetch cannot, yet the destination can still challenge or refuse it.\n\nThis has two consequences.\n\nFirst, the browser capability must report refusal as refusal. A rendered challenge page with status 200 is not the requested article. Success needs a content assertion: selector observed, expected heading present, schema satisfied, or another target-specific check.\n\nSecond, the system must not market Browser Run as a way around a publisher's controls. Robots rules, authorization, terms, rate limits and data handling remain part of the invocation policy. Browser execution changes the client. It does not confer permission.\n\nOne Hacker News commenter said they moved to remote browser rendering because bot protection made direct fetching unworkable. Another asserted that Perplexity was using Cloudflare Browser Rendering for scraping. Those are observations from named operators, not universal proof of bypass, permission, scale, reliability or present product behavior. They establish that practitioners reach for this category of tool in the exact gap between plain HTTP and executed pages. They do not settle whether any particular target should be fetched.\n\n## The output can be wrong even when the browser worked\n\nTransport success and representation correctness are separate gates. A GitHub report against `/crawl` showed root-relative image paths being resolved as page-relative paths in converted Markdown, producing broken image URLs while the HTML output remained correct. That is an externally reported converter defect on particular pages, not proof that all current Markdown is broken. It demonstrates why the receipt should retain the source URL, format, converter version when available, and a second representation for material captures.\n\nFor critical evidence:\n\n1. capture rendered HTML plus the representation used downstream;\n2. retain the final URL after redirects;\n3. hash both outputs;\n4. validate required links or fields against the HTML;\n5. label model-extracted JSON as derived;\n6. store a screenshot when the claim is visual;\n7. fail closed when a required selector or assertion is absent.\n\nThe `/json` action deserves an extra warning. Cloudflare documents it as AI-assisted extraction and says the default model is Workers AI's Llama 3.3 70B FP8 Fast unless another provider is supplied. A schema can constrain shape. It cannot make the content deterministic or true. JSON produced by a model is derived evidence and should preserve the prompt, schema, model identity, input hash and validation result. It should never overwrite the rendered source.\n\n| Evidence status | Browser example | What can be claimed |\n| --- | --- | --- |\n| observed | screenshot visibly contains an error banner | the banner was visible in that capture |\n| derived | model maps rendered page into a product schema | the model produced fields from that input |\n| specified | Cloudflare documents a request limit | the published contract states the limit |\n| implemented | catalogue row and adapter exist in code | this version contains the path |\n| deployed | production endpoint accepts the row | the deployed version exposes it |\n| reproduced | controlled call returns the expected representation | the tested input worked at that time |\n| externally attested | named operator reports a failure or use | that operator reported that experience |\n\n## Crawl is a queue, not a big page request\n\nThe beta `/crawl` action is asynchronous. A POST creates a job; subsequent reads retrieve status and results. Cloudflare says jobs may run for up to seven days and results remain available for fourteen days. That temporal shape belongs in the catalogue contract. A row that blocks a model turn until an entire crawl completes is the wrong projection.\n\nUse three capabilities instead:\n\n```text\nCRAWL_CREATE(url, limit, depth, formats) -> job_id receipt\nCRAWL_STATUS(job_id)                     -> progress receipt\nCRAWL_RESULTS(job_id, cursor)            -> bounded page of artifacts\n```\n\nThe catalogue can project those rows into an asynchronous REST API, terminal commands and model tools while retaining one authority policy and one lineage chain. Each result page should point back to the create receipt and catalogue snapshot.\n\nCloudflare's current Free limits specify five crawl jobs per day and one hundred pages per crawl. A March 2026 Hacker News comment multiplied those two numbers and questioned a 500-page daily ceiling. That is a reasonable reading of the Free limits now published, but the commenter described the documentation they saw and framed the concern more broadly. It is not independent evidence of a paid-plan cap. The current limits page says paid defaults can be increased and does not list the same crawl-specific table under Paid. The article therefore narrows the anecdote instead of repeating it as a current universal limit.\n\nA second operator built a two-script, zero-dependency CLI covering all nine REST endpoints then documented, including `/crawl`. That externally attests that the REST surface was usable as a coherent toolset for one builder. It does not prove our adapter, our credentials or today's endpoint. Our own proof remains the measured `/markdown` receipt above.\n\n## The authority boundary is larger than the URL\n\nA browser can send cookies, custom headers, HTTP credentials and injected scripts. It can follow redirects to a different host, load subresources from many hosts, download data, and execute code supplied by the destination. Treating authority as an allowlist on the initial URL is inadequate.\n\nThe minimum policy envelope includes:\n\n| Boundary | Required control |\n| --- | --- |\n| scheme | allow `https:`; reject `file:`, `data:`, local protocols |\n| destination | resolve DNS and reject private, loopback, link-local and metadata addresses |\n| redirects | revalidate every redirect target |\n| subresources | block or constrain hosts when the task permits |\n| credentials | declare exactly which cookies, headers or HTTP auth may leave |\n| scripts | prohibit untrusted catalogue rows from injecting code |\n| downloads | disable or quarantine with size and type limits |\n| duration | bounded navigation, selector and overall operation timeouts |\n| wallet | per-invocation browser-ms budget and caller quota |\n| output | byte limit, format validation, hashing and secret scan |\n\nThis is SSRF defense and denial-of-wallet defense in one place. The model should never receive a raw “browse any URL with these headers” primitive when a narrower row can express the job. A malicious directory row must not be able to expand its own host authority, supply a metadata address, or ask the adapter to return cookies in the receipt.\n\nThe receipt should be useful without becoming a credential leak:\n\n```json\n{\n  \"capability_key\": \"BROWSER_MARKDOWN\",\n  \"catalogue_version\": \"sha256:…\",\n  \"requested_url\": \"https://example.com\",\n  \"final_url\": \"https://example.com/\",\n  \"authority_policy\": \"public-docs-v3\",\n  \"completion\": {\"kind\": \"heading\", \"observed\": true},\n  \"format\": \"markdown\",\n  \"http_status\": 200,\n  \"browser_ms_used\": 207.706,\n  \"elapsed_ms\": 1418,\n  \"body_sha256\": \"sha256:…\",\n  \"credentials\": {\"token\": \"redacted\", \"cookies_sent\": false}\n}\n```\n\nReplay means invoke the same capability version with the same public inputs and policy, then compare receipts. It does not mean persist and resend an expired token. Repair means change the row or adapter under review—perhaps the selector, format, timeout or allowed host—then issue a new catalogue version and preserve the failed receipt. The ledger makes failure part of lineage instead of rewriting history.\n\n## REST for bounded transforms; Puppeteer for interaction\n\nQuick Actions cover common one-shot representations with smaller contracts. Puppeteer or Playwright is appropriate when the task genuinely requires interaction across states: click, type, authenticate, paginate, reuse a session, inspect requests, or coordinate multiple pages.\n\n| Requirement | Prefer | Reason |\n| --- | --- | --- |\n| one URL to Markdown | Quick Action | bounded request and direct browser-time receipt |\n| screenshot plus HTML | `/snapshot` | representations share one capture |\n| named CSS fields | `/scrape` | selector contract is explicit |\n| multi-page site collection | `/crawl` | asynchronous job semantics |\n| click through a flow | Puppeteer/Playwright | stateful interaction |\n| persistent authenticated workspace | reusable session | cookies and state are intentional |\n| stable public JSON endpoint | ordinary `fetch()` | no browser evidence is needed |\n\nThe decision can be mechanized in the canonical catalogue. Discovery exposes the specific transform first. Authority hides session tools from callers that do not need credentials or interaction. Invocation validates URL and completion conditions. Receipts normalize REST and session results into the same lineage fields. Repair can replace an implementation without changing the capability's public meaning.\n\nThis is where the system claim becomes concrete. One catalogue row drives discovery text, input schema, authority, adapter selection, receipt shape, replay, repair documentation, model-tool projection, CLI help and admin controls. The browser is not a second architecture. It is one implementation family behind the catalogue.\n\n## What the operator reports change—and what they do not\n\nThe people-source set is deliberately mixed.\n\n- A Cloudflare engineer reported `browser.close()` hanging in local Vite and Wrangler development while production succeeded. This supports testing local and deployed lifecycle separately.\n- A user reported REST error codes 7003 and 7000 despite a token and account identifier they had verified. This supports returning Cloudflare's structured error body and the chosen endpoint in the receipt; it does not prove the present API is generally misconfigured.\n- A crawl user reported malformed root-relative image URLs in Markdown while HTML stayed correct. This supports cross-format validation.\n- A commenter questioned crawl throughput based on the published limit arithmetic. This supports showing the limit calculation and reading the current plan table, not a universal paid-plan conclusion.\n- A CLI author reported exercising the full REST family. This supports the coherence of Quick Actions as a practical interface for that author.\n- Two other commenters described using or observing Browser Rendering for scraping and Markdown distillation. These support the use case, not permission, bypass success, commercial scale or adoption.\n\nOperator evidence is valuable here because it reveals failure modes absent from a happy-path reference: lifecycle hangs, auth-shaped errors, converter defects and throughput surprises. It remains externally attested evidence. The specification defines the contract; the fresh receipt establishes what this build reproduced; operator reports tell us which edges deserve tests.\n\n## The operating rule\n\nUse Browser Run when the thing you need does not exist until a browser executes the page, or when the required artifact is browser-specific. Name the representation. Name the completion condition. Constrain the authority. Meter browser time. Preserve the source alongside every derived form.\n\nDo not call it a bypass. Do not call model-shaped JSON fact. Do not call one successful render availability. Do not call a historical issue a current universal defect.\n\nWhen those boundaries are encoded once in the capability catalogue, the same browser operation can be discovered by a model, invoked from a terminal, projected as an API, receipted in the ledger, replayed after a change and repaired without losing its history. That—not remote Chrome by itself—is what makes Browser Rendering part of an operating system.","register":"essay","tags":["cloudflare","cloudflare-os","browser-run","browser-rendering","puppeteer","web-scraping","evidence","receipts","security"],"style":{},"claims":[{"id":"c1","text":"Browser Run can execute a page and return a browser-specific representation; it should be used only when that execution or representation is material.","section":"Browser Rendering is an evidence adapter, not a better fetch()","tier":"system","source_ids":["s1","s2","s4"],"why_material":"It defines the narrow boundary between ordinary HTTP and browser execution."},{"id":"c2","text":"The current Quick Actions surface exposes ten representation-oriented actions, including beta crawl.","section":"The endpoint is a choice about evidence","tier":"fact","source_ids":["s1","s2","s3","s4","s5","s6","s7","s8"],"why_material":"A caller can request the exact evidence form instead of receiving a generic browser primitive."},{"id":"c3","text":"Snapshot can capture two or more formats from one request, including content, screenshot, Markdown and the accessibility tree.","section":"The endpoint is a choice about evidence","tier":"fact","source_ids":["s3"],"why_material":"Critical captures can preserve visual and structural evidence from the same browser state."},{"id":"c4","text":"Default page-load completion can return empty or incomplete SPA results, so a browser capability needs a target-specific completion assertion.","section":"A browser can execute a page without proving the page is complete","tier":"system","source_ids":["s4","s5"],"why_material":"HTTP success and browser execution do not prove application completion."},{"id":"c5","text":"This build reproduced /markdown against example.com with status 200, 207.706 browser milliseconds and 1,418 ms client elapsed time.","section":"The first-party receipt: 208 browser milliseconds, not a speed claim","tier":"measurement","source_ids":["r1","s2","s9"],"why_material":"It is a fresh bounded receipt, not a statement that the adapter would work."},{"id":"c6","text":"Quick Actions are billed for browser time, while direct Browser Sessions add session-concurrency pricing.","section":"The bill is browser time, and sessions add a second meter","tier":"fact","source_ids":["s9"],"why_material":"The two invocation families have different cost surfaces."},{"id":"c7","text":"Workers Free includes ten browser minutes per day; Workers Paid includes ten hours per month and lists $0.09 for each additional browser hour.","section":"The bill is browser time, and sessions add a second meter","tier":"fact","source_ids":["s10","s9"],"why_material":"Browser time must be metered at invocation and account levels."},{"id":"c8","text":"Browser sessions should be explicitly closed and observed because idle sessions consume time; one historical issue reported close hanging in local development while production worked.","section":"The bill is browser time, and sessions add a second meter","tier":"anecdotal","source_ids":["p2","s10","s11"],"why_material":"Lifecycle defects can become denial-of-wallet defects."},{"id":"c9","text":"Cloudflare says a custom User-Agent does not bypass bot protection and Browser Run requests remain identifiable as bots.","section":"Cloud browser does not mean bypass","tier":"fact","source_ids":["s6"],"why_material":"Remote browser execution does not confer access or permission."},{"id":"c10","text":"Two operator comments describe using or observing Browser Rendering for scraping or Markdown distillation, but neither proves universal bypass, permission, scale or reliability.","section":"Cloud browser does not mean bypass","tier":"anecdotal","source_ids":["p5","p7"],"why_material":"The use case is real while the conclusion remains deliberately narrow."},{"id":"c11","text":"A crawl user reported root-relative image URLs being converted incorrectly in Markdown while the HTML remained correct.","section":"The output can be wrong even when the browser worked","tier":"anecdotal","source_ids":["p4"],"why_material":"Representation integrity must be checked independently of transport success."},{"id":"c12","text":"The JSON action uses model-assisted extraction, so its output is derived evidence even when it satisfies a schema.","section":"The output can be wrong even when the browser worked","tier":"system","source_ids":["s7"],"why_material":"Schema validity cannot be promoted into source truth."},{"id":"c13","text":"Crawl is asynchronous, may run for seven days and retains completed result data for fourteen days.","section":"Crawl is a queue, not a big page request","tier":"fact","source_ids":["s8"],"why_material":"Create, status and result retrieval need distinct capability contracts."},{"id":"c14","text":"The current Free crawl table lists five jobs per day and one hundred pages per crawl; an operator's broader throughput concern is not evidence of the present Paid limit.","section":"Crawl is a queue, not a big page request","tier":"system","source_ids":["p1","s10"],"why_material":"It preserves the useful arithmetic without universalizing a historical reading."},{"id":"c15","text":"One operator reported building a zero-dependency CLI across the then-documented REST endpoint family.","section":"Crawl is a queue, not a big page request","tier":"anecdotal","source_ids":["p6"],"why_material":"It externally attests coherent practical use of the REST surface for that builder."},{"id":"c16","text":"Browser requests can carry cookies, custom headers and HTTP credentials, so the authority boundary includes redirects, subresources, credentials, scripts, duration, wallet and output.","section":"The authority boundary is larger than the URL","tier":"system","source_ids":["s12","s8"],"why_material":"A host allowlist on the initial URL alone does not prevent SSRF, leakage or wallet abuse."},{"id":"c17","text":"One operator reported persistent 7003 and 7000 errors despite an account identifier and token they had verified.","section":"What the operator reports change—and what they do not","tier":"anecdotal","source_ids":["p3"],"why_material":"Structured upstream error bodies and endpoint identity belong in failure receipts."},{"id":"c18","text":"In this build, one canonical directory row supplies the capability contract, dispatch invocation and ledger lineage rather than creating a separate browser architecture.","section":"REST for bounded transforms; Puppeteer for interaction","tier":"implementation","source_ids":["s13"],"why_material":"The browser family must preserve the one-catalogue-to-all-projections system claim."}],"sources":[{"id":"s1","type":"specification","url":"https://developers.cloudflare.com/browser-run/quick-actions/","title":"Browser Run Quick Actions overview","quote":"Quick Actions are a set of REST API endpoints that provide common browser automation tasks.","summary":"Defines the bounded REST actions and their shared invocation model.","publisher":"Cloudflare","claim_ids":["c1","c2"]},{"id":"s2","type":"specification","url":"https://developers.cloudflare.com/browser-run/quick-actions/markdown-endpoint/","title":"/markdown — Extract Markdown from a webpage","quote":"The `/markdown` endpoint retrieves a webpage's content and converts it into Markdown format.","summary":"Defines the Markdown transform, REST and binding paths, authentication, and load controls.","publisher":"Cloudflare","claim_ids":["c1","c2","c5"]},{"id":"s3","type":"specification","url":"https://developers.cloudflare.com/browser-run/quick-actions/snapshot/","title":"/snapshot — Capture multiple page formats","quote":"The `/snapshot` endpoint combines multiple formats into a single request, so you do not need to call each endpoint separately.","summary":"Defines the multi-representation capture and its content, screenshot, Markdown and accessibility-tree formats.","publisher":"Cloudflare","claim_ids":["c2","c3"]},{"id":"s4","type":"specification","url":"https://developers.cloudflare.com/browser-run/quick-actions/content-endpoint/","title":"/content — Fetch rendered HTML","quote":"For JavaScript-heavy pages or Single Page Applications (SPAs), the default page load behavior may return empty or incomplete results.","summary":"Defines rendered HTML output and warns that default page completion can precede application completion.","publisher":"Cloudflare","claim_ids":["c1","c2","c4"]},{"id":"s5","type":"specification","url":"https://developers.cloudflare.com/browser-run/quick-actions/accessibility-tree-endpoint/","title":"/accessibilityTree — Capture the accessibility tree","quote":"For faster responses, advanced users can use `waitForSelector` to wait for a specific element instead of waiting for all network activity to stop.","summary":"Defines semantic roles, names and states and documents selector-based completion.","publisher":"Cloudflare","claim_ids":["c2","c4"]},{"id":"s6","type":"specification","url":"https://developers.cloudflare.com/browser-run/quick-actions/scrape-endpoint/","title":"/scrape — Scrape HTML elements","quote":"The `userAgent` parameter does not bypass bot protection. Requests from Browser Run will always be identified as a bot.","summary":"Defines selector extraction and the explicit non-bypass boundary.","publisher":"Cloudflare","claim_ids":["c2","c9"]},{"id":"s7","type":"specification","url":"https://developers.cloudflare.com/browser-run/quick-actions/json-endpoint/","title":"/json — Capture structured data using AI","quote":"In this example, Browser Run first calls Anthropic's Claude Sonnet 4 model.","summary":"Documents model-assisted schema extraction, model selection and fallback behavior.","publisher":"Cloudflare","claim_ids":["c12","c2"]},{"id":"s8","type":"specification","url":"https://developers.cloudflare.com/browser-run/quick-actions/crawl-endpoint/","title":"/crawl — Crawl web content","quote":"Crawl jobs have a maximum run time of seven days. If a job does not finish within this time, it will be cancelled due to timeout.","summary":"Defines asynchronous jobs, polling, seven-day runtime, fourteen-day result retention and browser parameters.","publisher":"Cloudflare","claim_ids":["c13","c16","c2"]},{"id":"s9","type":"specification","url":"https://developers.cloudflare.com/browser-run/pricing/","title":"Browser Run pricing","quote":"The `X-Browser-Ms-Used` header, which is returned in every Quick Actions response, reports browser time used for the request (in milliseconds).","summary":"Defines browser-hour and session-concurrency billing and the per-request usage header.","publisher":"Cloudflare","date":"2026-04-21","claim_ids":["c5","c6","c7"]},{"id":"s10","type":"specification","url":"https://developers.cloudflare.com/browser-run/limits/","title":"Browser Run limits","quote":"Crawl jobs per day | 5 per day","summary":"Current Free and Paid browser-time, request, session and crawl limits.","publisher":"Cloudflare","date":"2026-05-20","claim_ids":["c14","c7","c8"]},{"id":"s11","type":"specification","url":"https://developers.cloudflare.com/browser-run/puppeteer/","title":"Puppeteer on Browser Run","quote":"The `keep_alive` option sets the time in milliseconds to keep the browser open after the last activity.","summary":"Defines direct session control, binding use, session reuse and inactivity lifecycle.","publisher":"Cloudflare","claim_ids":["c8"]},{"id":"s12","type":"specification","url":"https://developers.cloudflare.com/browser-run/quick-actions/screenshot-endpoint/","title":"/screenshot — Capture a screenshot","quote":"Provide valid session cookies to access pages that require login:","summary":"Documents cookies, HTTP credentials, custom headers, viewport and load controls carried by a browser request.","publisher":"Cloudflare","claim_ids":["c16"]},{"id":"s13","type":"repository","url":"https://github.com/massoumicyrus/miscsubjects-architecture","title":"miscsubjects architecture","quote":"Tooling as data. The capability catalogue is rows in SQL, not tool schemas in a prompt.","summary":"Public architecture repository for the canonical catalogue and its projected client surfaces.","author":"Cyrus Massoumi","publisher":"GitHub","claim_ids":["c18"]},{"id":"p1","type":"hn","url":"https://hn.algolia.com/api/v1/items/47332926","title":"Cloudflare crawl endpoint","quote":"[0] seems to suggest even paid plans are effectively limited to 500 web pages per day, right?","summary":"Reads the Browser Rendering limits page and works out that 5 crawl jobs/day × 100 pages/crawl caps paid plans at roughly 500 pages per day. Negative — the throughput ceiling for anyone hoping to scrape at volume.","author":"gingerlime","publisher":"Hacker News","date":"2026-03-11","claim_ids":["c14"]},{"id":"p2","type":"github","url":"https://github.com/cloudflare/workers-sdk/issues/9945","title":"BUG: browser rendering browser.close() hangs","quote":"Like it says, `browser.close()` hangs in dev, in both vite and wrangler. It works fine in production, but this is a blocker for dev.","summary":"Minimal repro where browser.close() never returns under both vite and wrangler dev, though production is fine — filed by a Cloudflare engineer. Negative on Browser Rendering's local reliability.","author":"threepointone","publisher":"GitHub — cloudflare/workers-sdk","date":"2025-07-13","claim_ids":["c8"]},{"id":"p3","type":"github","url":"https://github.com/cloudflare/workers-sdk/issues/10864","title":"Cloudflare Browser Rendering API (Code 7003/7000) Failure in Worker","quote":"Despite verified API token permissions and Account ID, the API consistently returns a 400 error.","summary":"A content-fetcher Worker calling the Browser Rendering REST endpoint gets consistent 400s with error codes 7003/7000 despite verified token permissions and account ID. Second filing of the same problem by the same user after an earlier issue. Negative.","author":"face2611","publisher":"GitHub — cloudflare/workers-sdk","date":"2025-10-03","claim_ids":["c17"]},{"id":"p4","type":"github","url":"https://github.com/cloudflare/workers-sdk/issues/13406","title":"Browser Rendering /crawl API: Markdown converter incorrectly resolves root-relative image URLs","quote":"The `/crawl` endpoint's HTML-to-markdown converter incorrectly resolves root-relative image paths (paths starting with `/`) as page-relative paths, producing broken image URLs in the markdown output.","summary":"Crawled real sites with /crawl and found every root-relative image URL silently mangled in the markdown output while the html output is correct. The same converter bug was re-filed months later by another user as #14488. Negative — output correctness, not just uptime.","author":"matthew-sessions","publisher":"GitHub — cloudflare/workers-sdk","date":"2026-04-10","claim_ids":["c11"]},{"id":"p5","type":"hn","url":"https://hn.algolia.com/api/v1/items/44788890","title":"Perplexity is using stealth, undeclared crawlers to evade no-crawl directives","quote":"PS: perplexity is using cloudflare browser rendering to scrape websites","summary":"First-hand observation that Perplexity itself uses Cloudflare Browser Rendering for scraping — evidence the product is used at commercial scale for exactly the fetch-what-plain-HTTP-cannot-reach job. Neutral-positive.","author":"dhanushreddy29","publisher":"Hacker News","date":"2025-08-04","claim_ids":["c10"]},{"id":"p6","type":"hn","url":"https://hn.algolia.com/api/v1/items/47348398","title":"Cloudflare crawl endpoint","quote":"I built a CLI wrapper for the Browser Rendering REST API — covers all 9 endpoints including /crawl. Two Bun scripts, zero dependencies","summary":"Built and published a zero-dependency CLI covering all nine Browser Rendering REST endpoints (render, screenshot, PDF, scrape, AI extraction, multi-page crawl). Positive — someone exercising the full API surface in practice.","author":"nathanhouse","publisher":"Hacker News","date":"2026-03-12","claim_ids":["c15"]},{"id":"p7","type":"hn","url":"https://hn.algolia.com/api/v1/items/47572417","title":"ChatGPT won't let you type until Cloudflare reads your React state","quote":"hence i am just using cloudflare remote browser rendering.","summary":"Moving to a 'no browser' model where content is distilled to markdown pulled from Cloudflare's browser rendering rather than fetched directly, precisely because bot-protection makes plain fetch unworkable. Positive on the use case, sour on why it is necessary.","author":"lukewarm707","publisher":"Hacker News","date":"2026-03-30","claim_ids":["c10"]},{"id":"r1","type":"runtime_receipt","url":"https://miscsubjects.com/api/articles/cloudflare-os-browser","title":"Fresh first-party Browser Run /markdown receipt","quote":"Status 200; X-Browser-Ms-Used 207.706 ms; client elapsed 1418 ms.","summary":"Authenticated production Browser Run request against example.com; credential omitted; returned Markdown contained the expected heading.","publisher":"miscsubjects.com","date":"2026-07-26","claim_ids":["c5"]}],"prov":{"model":"Opus 5 (Claude Code)","action":"write"}}