Developers who use Claude Code no longer have to leave their terminal to check a web page. Anthropic has added a built-in browser to the desktop version of its coding agent, letting the AI read documentation, pull up API references and operate web-based development tools without opening Chrome, according to Digital Trends, which tested the feature. The publication said Claude Code can now “browse the web without opening Chrome.”
The browser is integrated into the agent’s workflow rather than bolted on as a separate window. When a task requires information from the web, Claude can navigate to a page, read its contents and incorporate what it finds into its next step. That matters for the mundane parts of programming that still send developers to a browser tab: checking a library’s documentation, finding the correct endpoint for an API, reading a configuration guide, or working through the pages of a web-based admin console. In a demonstration cited by Digital Trends, the agent used the browser to look up a function signature, adjusted its code accordingly and moved on without any handoff to a human.
9to5Mac, which also covered the release, said the addition is another sign that AI coding tools are moving beyond code completion toward complete development environments. The first wave of AI coding products suggested the next token. The current wave is expected to plan a change, make it, run the tests and verify the result. A browser is what lets an agent do that last part against the real world, instead of guessing from what it learned during training.
Claude Code, launched by Anthropic in February 2025, has become one of the most widely used AI coding tools among professional developers, competing with OpenAI’s Codex CLI, Google’s Gemini CLI and a crowded field of startups. Its early appeal was speed: it runs in the terminal, reads a repository and can carry out multi-step tasks with minimal prompting. The browser extension answers a different complaint, that agents are blind to the sources developers actually consult. Documentation lives on the web, and so does much of the feedback loop of modern software work.
The capability is part of a broader shift in how Anthropic describes the product. The company has talked about agents that not only write code but also manage the surrounding work of software development: reading tickets, updating documentation, deploying services and responding to the feedback those systems produce. Web access is the missing piece that ties those activities together, analysts said, because so much of that work happens in browsers rather than in code files. A developer fixing a broken integration, for example, often ends up reading a vendor’s status page; an agent with a browser can do that reading itself.
Security is the obvious counterweight. An agent with a browser can click on things, and an agent that can click on things can make mistakes or be steered toward them. Anthropic said browser actions run under the same permission and sandboxing model as the rest of Claude Code, and that users can approve or block navigation as they see fit. Enterprise customers, who are the product’s main revenue source, will want audit trails of what the agent visited and did, and Anthropic said those controls are part of the rollout. The company has also said it applies the same filters to what the agent reads online as it does to what users paste into a prompt.
Competitors are expected to follow. OpenAI has been adding tool use to its coding agents, and Google’s Gemini CLI already has access to the search and browsing infrastructure of its parent company. The difference, analysts said, is less about who ships first than about how reliably the browser works inside an agent loop: whether it can recover from a dead link, handle pages that render with JavaScript, and stay on task when the web disagrees with its assumptions.
The release also highlights how quickly coding tools are becoming the proving ground for agentic AI, the family of products that act on instructions rather than just answering them. Anthropic, OpenAI and Google have all described software development as the most commercially promising use of agents, and browser access is a capability they are racing to make reliable. The differences between their products now come down to execution details that developers will discover only by using them.
For developers, the practical change is small and immediate: fewer trips out of the terminal, fewer context switches, less time spent copying text between windows. The larger change is architectural. Software development has always been an activity that happens partly in an editor and partly everywhere else. Anthropic’s browser is an acknowledgment that an AI coding tool, to be complete, has to live in both places at once.
