--- title: Anthropic's browser tool is now generally available (and... description: Anthropic has made Computer Use, the Browser tool, the Skills API, and the Files API generally available on the Claude Platform. The browser tool doesn't host... canonical: https://daily.dev/posts/anthropic-s-browser-tool-is-now-generally-available-and-it-works-differently-than-you-d-think--xmspdt6ef twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Anthropic's browser tool is now generally available (and it works differently than you'd think) | daily.dev og:description: Anthropic has made Computer Use, the Browser tool, the Skills API, and the Files API generally available on the Claude Platform. The browser tool doesn't host... og:url: https://daily.dev/posts/anthropic-s-browser-tool-is-now-generally-available-and-it-works-differently-than-you-d-think--xmspdt6ef og:image: https://api.daily.dev/og/posts/XMsPdt6EF.png og:image:alt: Anthropic's browser tool is now generally available (and it works differently than you'd think) og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Anthropic's browser tool is now generally available (and it works differently than you'd think) **[Collections](https://daily.dev/sources/collections)** · 3 min read · 18 upvotes · 2 comments ## Summary Anthropic has made Computer Use, the Browser tool, the Skills API, and the Files API generally available on the Claude Platform. The browser tool doesn't host or control a browser itself; instead Claude interacts with pages via structured element references (like ref_3) pulled from the accessibility tree, avoiding screenshot-and-coordinate guessing. Developers must host their own browser infrastructure. A new batching feature lets multiple browser actions be requested in a single model turn, cutting latency and cost, though error handling mid-sequence remains the developer's problem. The default toolset exposes 27 operations and adds roughly 6,600 input tokens per request before screenshots or accessibility data. Anthropic recommends running the browser in an isolated, minimally privileged container since Claude remains vulnerable to prompt injection and unexpected redirects. ## Content Anthropic has pushed several things out the door at once: a free learning platform called Claude Academy, and general availability of the browser tool, Skills API, and Files API on the Claude Platform. ## Claude Academy Claude Academy offers structured courses aimed at different experience levels. Claude 101 is for newcomers, there's an AI Fluency track focused on delegation skills, and more advanced courses cover APIs, coding, and building agents. Everything is free, you can browse without signing up, and completed courses earn certificates and badges. ## Browser tool The browser tool is worth understanding carefully, because it works differently than you might expect. Claude doesn't actually run a browser itself. Instead, it interacts with web pages through structured element references (like `ref_3`) pulled from the accessibility tree, rather than guessing pixel coordinates from screenshots. Developers access it via `browser_toolset_20260801` in the Claude API, host the browser themselves, and translate Claude's requests into actual actions. A new batching feature lets Claude request multiple browser actions in a single model turn, which reduces back-and-forth calls and cuts latency and cost. The executor still has to run actions in order and handle failures mid-sequence, so it's not magic, but it's a real improvement. The default toolset exposes 27 browser operations and adds roughly 6,600 input tokens per request before any screenshots or accessibility tree data. One thing Anthropic is upfront about: Claude remains vulnerable to prompt injection and unexpected redirects. They recommend running the browser in an isolated, minimal-access container. ## Files API The Files API is now generally available and lets you upload reference documents once rather than pasting them into every request. Anthropic claims it cuts task round trips by 20-40%. That said, a hands-on test comparing the Files API against pasting and against prompt caching tells a more nuanced story. Using a fake API-support-bot scenario with claude-sonnet-5 across five requests, the Files API actually billed *slightly more* than pasting (15,371 vs 15,246 input tokens). Prompt caching, by contrast, cut billed input to roughly a third by charging for the document once as a cache write and reading it cheaply afterward. Accuracy was identical across all three approaches. The conclusion: the Files API solves a document management problem, not a cost problem. If you're uploading the same reference material repeatedly, prompt caching is what actually reduces spend. The good news is you can combine both. ## Skills API The Skills API is also now generally available. It's designed to support Claude Managed Agents, letting developers build things like personal finance helpers where Claude can call predefined capabilities rather than improvising from scratch each time. ## Questions this post answers ### How does Claude's browser tool interact with web pages instead of using screenshots? Claude's browser tool uses structured element references, like ref_3, pulled from the accessibility tree rather than taking screenshots and guessing pixel coordinates. This lets Claude reference an actual page element directly instead of clicking roughly where a button appears to be. Developers access it via the Claude API using browser_toolset_20260801 but must host the browser and execute Claude's requested actions themselves. _Teams wiring up Claude's browser tool can track API changes like this one on daily.dev._ ### How much does enabling Claude's default browser toolset add to my token usage? The default browser toolset exposes 27 operations and adds around 6,600 input tokens to every request, before any screenshots or accessibility tree data are included. This overhead matters for anyone running browser-enabled Claude calls at scale, since it applies just from having the toolset available, regardless of how many actions are taken. _Anyone budgeting AI agent costs can follow token-overhead details like this on daily.dev._ ### What does batching multiple browser actions in one Claude model turn actually solve? Batching lets multiple browser actions be requested in a single model turn instead of round-tripping to the model after every click and scroll, cutting the latency and cost that build up when each action needs a fresh model call. It doesn't eliminate complexity, though: the executor still has to run actions in order and handle failures if, say, action 3 of 7 breaks due to an unexpected redirect. _Developers designing agent executors can follow batching trade-offs like these on daily.dev._ ## Community discussion Top comments from developers on daily.dev. **@taiwofrancis** · 2 upvotes > The structured element references are actually a pretty clever approach. Less “AI guessing where to click,” more “AI actually knowing what it’s interacting with.” **@baz14** · 0 upvotes > The internet was originally designed for **people**, then for **apps**. Today, a new type of client has emerged - **Al agents**🤚The web must now be adapted specifically for them, creating a new type of interaction. MCP is one of the protocols developed to standardize this interaction between Al systems and external tools and services. But slow adoption limits/kill progress🙁 > > Browser tool are, in part, a reaction to the lack of agent-native interfaces and proper MCP integration. Also, MCP itself is still under active development. --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#automation](https://daily.dev/tags/automation), [#claude](https://daily.dev/tags/claude), [#anthropic](https://daily.dev/tags/anthropic), [#prompt-injection](https://daily.dev/tags/prompt-injection) [View this post on daily.dev](https://daily.dev/posts/anthropic-s-browser-tool-is-now-generally-available-and-it-works-differently-than-you-d-think--xmspdt6ef) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Anthropic's browser tool is now generally available (and it works differently than you'd think)","url":"https://daily.dev/posts/anthropic-s-browser-tool-is-now-generally-available-and-it-works-differently-than-you-d-think--xmspdt6ef","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/anthropic-s-browser-tool-is-now-generally-available-and-it-works-differently-than-you-d-think--xmspdt6ef"},"datePublished":"2026-08-21T16:56:43.268Z","dateModified":"2026-08-27T13:06:00.990Z","description":"Anthropic has made Computer Use, the Browser tool, the Skills API, and the Files API generally available on the Claude Platform. The browser tool doesn't host...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/f56c6585136e3237ce6dd8ae1f426fe9?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/f56c6585136e3237ce6dd8ae1f426fe9?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":2,"discussionUrl":"https://daily.dev/posts/anthropic-s-browser-tool-is-now-generally-available-and-it-works-differently-than-you-d-think--xmspdt6ef","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":18},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":2}],"keywords":"ai-agents,automation,claude,anthropic,prompt-injection","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Anthropic's browser tool is now generally available (and it works differently than you'd think)"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/anthropic-s-browser-tool-is-now-generally-available-and-it-works-differently-than-you-d-think--xmspdt6ef","comment":[{"@type":"Comment","text":"The structured element references are actually a pretty clever approach. Less “AI guessing where to click,” more “AI actually knowing what it’s interacting with.”","datePublished":"2026-08-22T09:55:09.740Z","url":"https://daily.dev/posts/XMsPdt6EF#c-lxPPDk2PD","author":{"@type":"Person","name":"Taiwo Francis Oguntade","url":"https://daily.dev/taiwofrancis","image":"https://media.daily.dev/image/upload/s--px3_Pvo4--/f_auto/v1785515368/avatars/avatar_FStpm3ZejUD3qGlEeoOUS?_a=BAMAMicg0"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2}},{"@type":"Comment","text":"The internet was originally designed for people, then for apps. Today, a new type of client has emerged - Al agents🤚The web must now be adapted specifically for them, creating a new type of interaction. MCP is one of the protocols developed to standardize this interaction between Al systems and external tools and services. But slow adoption limits/kill progress🙁\nBrowser tool are, in part, a reaction to the lack of agent-native interfaces and proper MCP integration. Also, MCP itself is still under active development.","datePublished":"2026-08-23T07:12:21.246Z","url":"https://daily.dev/posts/XMsPdt6EF#c-9ZsuydNnq","author":{"@type":"Person","name":"Baz","url":"https://daily.dev/baz14","image":"https://lh3.googleusercontent.com/a/ACg8ocKd2iFU8YS4sV-rBJptuYSF8EHeQnm23N_dcXE1uON7_GMIlZg=s96-c"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/anthropic-s-browser-tool-is-now-generally-available-and-it-works-differently-than-you-d-think--xmspdt6ef#faq","mainEntity":[{"@type":"Question","name":"How does Claude's browser tool interact with web pages instead of using screenshots?","acceptedAnswer":{"@type":"Answer","text":"Claude's browser tool uses structured element references, like ref_3, pulled from the accessibility tree rather than taking screenshots and guessing pixel coordinates. This lets Claude reference an actual page element directly instead of clicking roughly where a button appears to be. Developers access it via the Claude API using browser_toolset_20260801 but must host the browser and execute Claude's requested actions themselves. Teams wiring up Claude's browser tool can track API changes like this one on daily.dev."}},{"@type":"Question","name":"How much does enabling Claude's default browser toolset add to my token usage?","acceptedAnswer":{"@type":"Answer","text":"The default browser toolset exposes 27 operations and adds around 6,600 input tokens to every request, before any screenshots or accessibility tree data are included. This overhead matters for anyone running browser-enabled Claude calls at scale, since it applies just from having the toolset available, regardless of how many actions are taken. Anyone budgeting AI agent costs can follow token-overhead details like this on daily.dev."}},{"@type":"Question","name":"What does batching multiple browser actions in one Claude model turn actually solve?","acceptedAnswer":{"@type":"Answer","text":"Batching lets multiple browser actions be requested in a single model turn instead of round-tripping to the model after every click and scroll, cutting the latency and cost that build up when each action needs a fresh model call. It doesn't eliminate complexity, though: the executor still has to run actions in order and handle failures if, say, action 3 of 7 breaks due to an unexpected redirect. Developers designing agent executors can follow batching trade-offs like these on daily.dev."}}]} ```