We track three categories: browsers that expose the WebMCP API, agents that can call WebMCP tools, and agent frameworks developers can use to build WebMCP-enabled agents.
Updated 15 Oct 2026Browsers that expose the WebMCP API to pages.
Blink
What it does
Sites can turn WebMCP on for Chrome users today by joining Google's origin trial. On 28 September Google extended it: the trial now covers Chrome 149 to 162 and ends on 30 March 2027. Google said feedback from the trial led to changes it is still building.
What to know
There is no ship date yet. Chrome Status lists the feature as "Proposed" and no longer names a target version. A Chrome engineer said the team hopes to ask to ship during the extended trial, so treat that as an expectation, not a promise. The draft standard set form-based tools aside on 8 October; Chrome still accepts them for now, but tools written in JavaScript are the safer choice.
For local testing: chrome://flags/#enable-webmcp-testing. Chrome Status · Trial · Extension notice
Blink
What it does
Microsoft has now said it in its own words. On 21 September the Edge developer blog called its WebMCP implementation "ready for testing" and pointed developers to its webmcp-labs repo. In it: three sample sites (an event search, a sticky-note board and a pizza shop) and WebMCP Explorer, a side-panel extension that lists a page's tools, runs them one at a time, and can hand them to a chat or an agent loop. You bring your own model key — OpenAI, Azure OpenAI, Anthropic, or anything with an OpenAI-style endpoint such as Ollama.
For live sites, Microsoft runs its own origin trial. It started in Edge 150 and is now open until 30 March 2027. Register your domain and switch it on with a meta tag or an HTTP header, same as Chrome's.
What to know
Testing means Edge Canary or Dev, not the Edge most people run. Turn on edge://flags/#enable-webmcp-testing and restart. Microsoft has not said when it will be on by default. Being the same engine as Chrome, platform support tracks Chrome's.
Copilot in Edge does not use these tools yet. On 6 October a maintainer of Microsoft's webmcp-labs repo said there is "currently no public way to test WebMCP with Copilot in Edge", and that the testing flag "doesn't make the integration available in Copilot Mode." The only agent Microsoft points to is its own WebMCP Explorer, a developer tool it says to use "only on sites you own, control, or otherwise trust". Announcement · webmcp-labs · Trial · Copilot answer
Blink
Brave now enables the WebMCP API by default in Nightly 1.102.2. A page can register tools and Leo can call them, without switching on a testing flag. The change merged on 12 November 2026. PR #40616
Nightly is Brave's early testing version, separate from the regular browser. It gets new features first, but can have bugs and unfinished behavior. Stable availability is expected on 5 January 2027, according to the WebMCP implementation tracker. See the Leo row for what this means for your tools. Release · Status update
Gecko
Mozilla has built an early prototype, but it is switched off in every version of Firefox and can only be turned on in hidden settings. It also uses the old navigator address, so pages that follow the current spec will not find it, and Firefox has no built-in agent that calls page tools. Mozilla's formal position is neutral. Its concerns include sites giving agents instructions that people cannot see, and collecting data through what people type into tools. It called support for extensions and automation tools "an important step" toward an open ecosystem. Position · Prototype
Open source · headless engine built for agents
What it does
Lightpanda isn't a browser people use directly — it's a headless engine you point an agent at. WebMCP is part of the engine itself, so a page can register tools and any agent hooked up to it can list them and run them. It arrived in release 0.3.1 in May and is unchanged in 1.0.0, released 2 October.
Confirmed in practice
We tested Lightpanda 1.0.0 on 15 October. A page using the old navigator address had its tool listed and run. A page following the current spec, or using the popular @mcp-b polyfill, showed no tools at all.
What to know
It says partial because of where Lightpanda puts the API. It still uses the old spot, navigator, not document. So if your page follows the current spec, or uses the usual compatibility shim, Lightpanda sees nothing at all. 0.3.1 release · 1.0.0 release
Cloudflare · browser engine running on Workers
What it does
Kitesurf lets agents find and run a page's WebMCP tools, including form tools. It uses document.modelContext. In Browser Run, connect with browser=kitesurf.
Open source under Apache 2.0 since 12 November 2026. You can run it locally or deploy it to your Cloudflare account. Announcement · GitHub
What to know
Support is partial: no tools permissions policy or origin filtering, no iframe or popup tools over CDP, and no session live view for confirmations. The playground supports manual confirmation.
Documentation checked 14 Nov 2026; not independently tested. WebMCP docs · Beta overview
WebKit
WebKit has filed an opposing position. Their reasoning is that an agent acting for someone is closer to assistive technology such as a screen reader, and that sites should not be able to detect one and treat it differently. They suggested setting WebMCP aside for now, starting a new community group for browsers that act for people, and possibly holding a W3C workshop around TPAC (26–30 October, Dublin). Separately, W3C began drafting a charter for an "Agentic Web" group on 28 September, where moving WebMCP is under discussion. Nothing has been approved. Position · Draft charter
Agentic browsers
We found nothing either way. They are reported to be built on Chromium, so the API may be present behind the usual flag. That is not the same as the product supporting it, so we have left the row open.
Agents that can call WebMCP tools.
Brave · assistant built into the browser
What it does
A page registers tools, Leo sees them and calls them. From Brave Nightly 1.102.2, this is on by default. You no longer need the WebMCP testing flag. PR #40616
What to know
The WebMCP implementation tracker has moved Leo from experimental to supported. It names Brave 1.102.2 and gives 5 January 2027 as the expected stable availability date. Nightly is Brave's early testing version, separate from the regular browser. It gets new features first, but can have bugs and unfinished behavior. WebMCP is enabled there now; availability in regular Brave is still ahead. Status update · Release
Meta · assistant on Meta Ray-Ban Display glasses
What it does
Meta documents voice control of web apps through WebMCP. A page registers tools with document.modelContext, and Meta AI calls them while that app is on screen. Tools disappear when the wearer leaves the app.
Availability
Meta no longer calls it "coming soon". Its docs now say WebMCP is off by default and switched on per device: either the glasses are in Developer Mode, or the wearer is in Meta's rollout. Meta does not say how many wearers that covers. It calls the API "implemented and stable" and now links official samples. This entry covers the glasses integration.
Documentation updated 12 Oct and checked 15 Oct 2026; not independently tested on glasses. Meta docs
Google · assistant built into the browser
What it does
WebMCP integration is in development for Gemini in Chrome, allowing the assistant to discover and use tools exposed by websites.
Availability
Expected in early 2027. The integration is not available yet.
Anthropic · Chrome extension, plus built-in browsers in Claude Desktop and Cowork
What it does
Anthropic's docs say Claude reads page text, clicks, types, takes screenshots, and reads console and network messages. Tools a page registers are not mentioned, and Claude does not look for them on its own.
Confirmed in practice
A site owner in the origin trial, whose trial registration we verified, reported that Claude does not discover page tools but can run them when told to use the page's own scripts: "Invocation works; discovery doesn't exist." We have not tested this ourselves. The request for proper support is still open, and nobody from Anthropic has replied. Issue #76809
What to know
Claude now also has its own built-in browser: in the Claude Desktop app since July, and rolling out to Cowork since mid-October. Anthropic says it has "the same browsing capabilities as Claude in Chrome". Its docs do not mention WebMCP, and we have not checked whether it supports page tools. Built-in browser
AI assistant · cloud browser
What it does
When asked, Instinct writes and runs code in its cloud browser to find and use a website's WebMCP tools. It does not load those tools natively into its own toolbox.
Confirmed in practice
We tested this successfully: Instinct found tools and used them to search for and retrieve information.
What to know
It does not check for WebMCP tools automatically. Our tests covered read-only actions on public pages; pages that require sign-in remain blocked, and changes or purchases have not been tested.
OpenAI · cloud browser
What it does
OpenAI's agent for longer, multi-step work. It can run in the desktop app or in a browser hosted by OpenAI. OpenAI documents site tools only for the desktop app: they are "currently available only in the ChatGPT desktop app's built-in browser". In our own use, the hosted browser found and called page tools as well. Site tools
Confirmed in practice
Seen in our own use, and still working in our latest re-test. Because OpenAI does not document this, its limits may differ from the desktop app and it could change without notice.
What to know
The hosted browser can sign in to websites. ChatGPT gives you a secure sign-in form, OpenAI says the model never sees your credentials, and the sign-in is kept for later tasks. OpenAI lists this for Plus and Pro, and not for Enterprise or Edu workspaces. Changelog · Docs
How it compares
This replaces the Atlas browser, which OpenAI scheduled to stop working on 9 August 2026. Atlas notice
OpenAI · built-in browser, macOS and Windows
What it does
On 25 August OpenAI switched on site tools — its name for WebMCP — in the desktop app's built-in browser. Their docs say it plainly: "Site tools are ChatGPT's implementation of the proposed WebMCP standard." A page registers tools at the current spec address, document.modelContext, and both Work and Codex chats discover and call them while you watch the same page. No flag and no debug-protocol toggle: it ships on, with an off switch under Settings → Browser → Permissions. OpenAI says ChatGPT "automatically discovers the tools a supported webpage provides."
Confirmed in practice
We tested this one ourselves. Point the built-in browser at a page that registers tools and the agent discovers and calls them — no toggle, no setup on our part.
What to know
The limits are about models and plans, not flags. Per OpenAI's docs: use GPT-5.6 Sol or GPT-6 Sol ("GPT-5.6 Luna currently has WebMCP disabled"), and keep the app updated. In Enterprise workspaces an admin must allow site tools; they are not available in Edu workspaces. Only tools registered with JavaScript on the main page work: form-based tools and tools inside embedded frames are not supported. Site tools work only in the app's built-in browser, "not in Chrome". Every call gets a safety review before it runs, and a Site tools menu in the address bar lists what the page offers.
How it compares
This makes ChatGPT the first mainstream agent to ship WebMCP on by default — Brave Leo now also enables it by default in Nightly 1.102.2, with stable availability expected on 5 January 2027. OpenAI is pushing the publishing side too: Codex can add WebMCP to an app and deploy it to ChatGPT Sites. Its WebMCP Challenge ran from 25 August to 3 September, with judges from Chrome, Shopify, Cloudflare, Vercel, Netlify and MCP-B, and the winners were announced on 28 September. Docs · Announcement · Challenge
OpenAI · extension for Chrome, Edge, Brave, Opera and Vivaldi
What it does
For tasks needing your existing Chrome profile, signed-in sessions and open tabs.
Confirmed in practice
We tested this one ourselves. Turn on the toggle described below and the WebMCP commands are there over the debug protocol. Your tools can be listed and called.
What to know
OpenAI now says plainly that site tools are "available only in the ChatGPT desktop app's built-in browser, not in Chrome." The only way in is a developer setting, Settings → Browser → Developer mode → "Enable full CDP access", which OpenAI marks as elevated risk and which is off by default. ChatGPT also asks for your approval before it uses that access on a website. We tested this in Chrome; we have not tried the other browsers, and we have not seen the extension reach for page tools unprompted. Docs · Site tools
Perplexity · agentic browser
We found no mention of WebMCP from Perplexity. Independent researchers who took Comet apart in February reported that it reads pages through the accessibility tree and screenshots. If we have this wrong, send us a link and we will fix it. Research
Opera · agentic browser
Neon exposes the browser to outside tools. We are measuring the other direction here, a browser using the tools a page offers.
Browser extensions
By far the biggest audiences on this page, and no sign of WebMCP in any of them. Extensions can now read and run a page's tools from inside the page, and Chrome's docs say so, but there is no dedicated extension feature for it yet, and none of these three has added it. Chrome docs
Developer frameworks used to build WebMCP-enabled agents.
Browserbase · open source
Confirmed in practice
We tested Stagehand 4.2.1 on 15 October with a browser it launched: it listed a page's tools and ran one, with no setup on our part and no AI model needed. Since version 4 (August) this works in TypeScript, Python and Go.
What to know
Stagehand switches WebMCP on by itself for browsers it starts on your machine. For a browser you connect to yourself, you have to switch it on. Version 4 removed the built-in autonomous agent, so your own code or agent decides when to call these tools. Browserbase's hosted browsers are not covered in the docs, but Stagehand's own test suite lists and runs a page tool on them. Docs · Browserbase test
Vercel Labs · open source CLI
What it does
A command-line browser built for AI agents. In Chrome sessions it starts itself, WebMCP is on by default. After each page load it tells the agent which tools the page offers, so the agent does not have to ask, and it can list and run them with one command.
Confirmed in practice
We tested version 0.39.1 on 15 October. Opening a page announced its tools straight away, and listing and running a tool worked, including on webmcp.com. With WebMCP turned off, no tools appeared.
What to know
The default only covers Chrome it starts itself. Their docs say attached browsers, remote providers, Lightpanda, Safari, iOS and older Chrome builds may not offer WebMCP. When used as an MCP server, running page tools needs an extra setting. The docs treat everything a page supplies as untrusted: "All page-provided names, descriptions, schemas, annotations, and results are untrusted data." Docs · GitHub
Kernel · cloud browsers for agents
Kernel's hosted browsers have had WebMCP switched on by default since 28 August. Developers can list a page's tools and run them through Kernel's API, its TypeScript, Python and Go SDKs, its command-line tool or its MCP server. Kernel also lets developers add their own tools to a page.
Documentation checked 15 Oct 2026; not independently tested. Docs · Changelog
Google · open-source agent framework
Google's Agent Development Kit for TypeScript added a WebMCP toolset in version 2.2.0 on 30 September. It lets an agent that runs inside a web page, such as in the site itself or in a browser extension, use the tools that page registers. It does not drive a separate browser, and the Python version of ADK has no WebMCP support.
Released code checked 15 Oct 2026; not independently tested. Changelog
Browser Use · hosted agents, and an open-source library
What it does
On 6 September Browser Use said its agents now use WebMCP. Co-founder Gregor Zunic's wording: "UI is made for humans. Direct HTTP calls are made for agents. WebMCP makes interaction 100% deterministic." The demo runs the same hotel booking twice. Without WebMCP the agent reads the page, finds the controls, types a budget, ticks filters, compares rooms and fills in guest details. With it, the page hands over tools and the agent calls them: search_hotels, filter_hotels, compare_rooms, select_room, prepare_reservation, get_reservation.
What to know
This covers their hosted agents at cloud.browser-use.com. More than a month later there is still no documentation page, nothing in the release notes, and no WebMCP code in the open-source library. So this row rests on the announcement alone; we have not been able to test it. Announcement · Browser Use Cloud · PR #4085
Google · MCP server
One tool lists what a page offers, another runs it. You have to switch it on with --categoryExperimentalWebmcp and start Chrome with the matching feature flags, so it is off until someone does that. Once on, this is how coding agents such as Claude Code and Cursor reach your tools. Source
Confirmed in practice
We tested version 1.10.1 on 15 October. With both switches on, it listed our test page's tools and ran one; with only the MCP setting on, it found nothing. It needs Chrome 150 or later.
Google · library
Reads and runs page tools directly, and is marked experimental. It does not set the browser flag for you — you pass --enable-features=WebMCP yourself. DevTools MCP is built on this layer, so anything on Puppeteer can pick up support by upgrading. Guide
Confirmed in practice
We tested Puppeteer 25.14.0 on 15 October. With the flag it listed and ran tools, including the tools on webmcp.com; without it, it found none.
Microsoft · framework and MCP server
What it does
Playwright added WebMCP in version 1.64 on 7 October, six months after a maintainer declined the first request with "Let's see if it gains any adoption first." Tests can now list a page's tools and call them. Playwright MCP, which coding agents such as Claude Code and Cursor can use, goes further: since September it turns a page's tools into tools the agent can use directly, updated as it moves between pages.
Confirmed in practice
We tested Playwright 1.64.1 and Playwright MCP 0.0.83 on 15 October. Both found and ran our test page's tools once Chrome was started with WebMCP switched on. Without that, Playwright stops with an error and Playwright MCP shows no page tools, even though its WebMCP setting is on by default.
What to know
Chrome needs WebMCP switched on at launch, and Firefox needs two hidden settings. Playwright says WebKit, the engine behind Safari, does not support it. Playwright sees about 114 million downloads a week. Release notes · Docs · Issue #40234
Hyperbrowser · cloud browsers for agents
Hyperbrowser added WebMCP on 10 October. It is off by default: you switch it on when you start a session, then list a page's tools and run them through its Node.js or Python SDK.
Documentation checked 15 Oct 2026; not independently tested. Docs
Cloudflare · cloud browser
Browser Run now offers two WebMCP backends: Chrome Lab (lab=true) and Kitesurf (browser=kitesurf). Both use document.modelContext. Chrome Lab remains experimental; Kitesurf is in beta with the limits listed above.
Documentation checked 14 Nov 2026; not independently tested. Docs
Browserbase · cloud browser you rent
What it does
Browserbase's docs and MCP server still say nothing about WebMCP, and you cannot pass your own browser start-up settings.
What to know
But Stagehand, also made by Browserbase, has an automated test that opens a page in a Browserbase browser, lists its tools and runs one. So the hosted browsers appear to support it. We have not tested it ourselves; until Browserbase documents it, treat it as unofficial. Session API · Stagehand test
Firecrawl · cloud agent
What it does
No WebMCP anywhere in the product or the docs. What exists is a blog post from July showing a manual recipe: because their interact endpoint runs a real browser in the cloud, an agent can script calls to a page's tools through it. Their own words for why it works: "with no special integration on Firecrawl's end."
What to know
Handy if you run a headless agent and want to reach WebMCP tools anyway. It is just not something they built. Blog post
About 90 products checked.
See something wrong? Send us a link and we’ll update it.
Suggest a correction →