What Is WebMCP? A New Web Standard for AI Agents
The answer to the question of what WebMCP is: a new standard that makes your website accessible to AI agents. We explain how it works and how to set it up.
The question of what WebMCP is has become a hot topic for anyone curious about how AI agents will interact with websites. WebMCP is a new proposed web standard that allows your website to expose its own functionality as ready-made tools for AI agents. The goal is to let agents use sites through direct, structured commands instead of clicking around like a human.
In short:
- WebMCP is being developed within the W3C, with contributions from Google and Microsoft engineers.
- It was announced on February 10, 2026, and is being tested in Chrome as an origin trial.
- It provides a browser-native API called navigator.modelContext.
- It doesn't replace MCP; it addresses a different need.
What is WebMCP?
WebMCP is a web standard that allows your website's features to be exposed to AI agents as callable tools. This standard is being developed under the W3C's Web Machine Learning Community Group. The effort is being led by engineers from Google and Microsoft (Medium).
At its core, WebMCP makes a website "understandable" for AI agents. The site owner defines the actions on their page in a structured format. The agent reads these definitions and can then use the site directly. Instead of taking a screenshot and interpreting it, the agent now calls an actual tool.
WebMCP was officially announced on February 10, 2026 (VyomEdge). It's currently in early preview in Chrome. The standard is still maturing, but its core architecture has taken shape.
What is WebMCP used for?
WebMCP allows agents to use websites more reliably and quickly. In the traditional approach, an agent visually analyzes a page and tries to click on buttons. This approach is slow and error-prone. With WebMCP, the site defines its own tools directly.
At the heart of this is a browser-native API called navigator.modelContext. Through this API, websites can expose their features to AI agents as structured, callable tools (Medium). Instead of interpreting the page, the agent calls these tools directly.
WebMCP uses the browser's internal systems. This means communication between the client and the tool happens almost instantly. There's no need to wait for a round trip to a remote server (Chrome for Developers). This is an advantage that directly improves the user experience.
WebMCP offers two different APIs. The Declarative API allows standard operations to be defined directly in HTML forms. The Imperative API, on the other hand, is used for more complex, dynamic interactions that require running JavaScript (Chrome for Developers). Businesses can choose either one or both, depending on their needs.
What's the difference between WebMCP and MCP?
WebMCP is not an extension or a new version of MCP. The two address different needs (Chrome for Developers). Understanding this distinction correctly determines which technology to choose and when.
With MCP, you can build your own tool server and call that server from clients like Claude Desktop. WebMCP, on the other hand, makes your already-existing website accessible to AI agents by adding just a few lines of JSON (Yapay Zeka Sözlük). In other words, while MCP requires a separate server infrastructure, WebMCP adds a thin layer on top of your existing website.
To summarize briefly:
| Feature | MCP | WebMCP |
|---|---|---|
| Where it's used | Separate tool server | Existing website |
| Access | Desktop AI clients | In-browser agents |
| Setup | Server development | JSON/API integration |
| Communication speed | Depends on network latency | Nearly instant |
Those curious about the general logic of the Model Context Protocol can also draw connections with the representation learning approach in our Contrastive Language Models post. For the general framework of agent architectures, you can also check out our Google AX post.
Which browsers support WebMCP?
As of September 2026, WebMCP is accessible through an origin trial covering Chrome versions 149 through 156 (AccessibilityChecker). This means the standard hasn't been permanently released yet and is still in the testing phase.
Microsoft has also joined this race. Edge is running its own origin trial starting from Edge 150, which will end on November 17, 2026 (AccessibilityChecker). Support from both major browser makers is a positive signal for WebMCP's future.
On the Chrome side, the origin trial covers eight versions, from 149 to 156. Edge, meanwhile, is currently running a more limited test with just one version (150). This difference shows that Chrome is testing WebMCP earlier and more broadly.
How do you use WebMCP?
To try out WebMCP, you first need to set up your local development environment. The process works step by step as follows:
- Go to chrome://flags/#enable-webmcp-testing in your Chrome browser.
- Set the flag to "Enabled."
- Restart Chrome (Chrome for Developers).
- Use the navigator.modelContext API to define your site's tools.
- Choose whichever suits your needs: the Declarative or Imperative API.
If you want to use WebMCP on a live website, the process is a bit different. You need to join Chrome's origin trial program and add the token provided to you to your site (AccessibilityChecker). This token tells the browser that your site is authorized to use the experimental feature.
In practice, the easiest starting point is adding a few lines of JSON to your existing site. WebMCP lets you make an already-existing website accessible to AI agents this way (Yapay Zeka Sözlük). There's no need for a major architectural overhaul.
Frequently asked questions
Is WebMCP a new version of MCP?
No. WebMCP is not an extension or modified form of MCP. The two are designed for different scenarios; MCP requires you to set up a separate tool server, while WebMCP integrates into your existing website.
Do I need a paid tool to test WebMCP?
No. For local testing, it's enough to enable Chrome's chrome://flags/#enable-webmcp-testing option. If you want to use it on a live site, you need to sign up for Chrome's free origin trial program.
Which browsers support WebMCP?
It's currently available as an origin trial in Chrome versions 149-156. Microsoft Edge is also running its own origin trial starting from Edge 150. Support from other browsers hasn't been clarified yet.
Is WebMCP safe for enterprise sites?
WebMCP is still in early preview and origin trial stages. Before using it in enterprise projects, it makes sense to keep an eye on the standard's maturation and the progress of the W3C process.
Sources
WebMCP is a candidate standard with the potential to fundamentally change the way websites communicate with AI agents. Even though it's still in an early stage, businesses need to keep a close eye on this development. At EngerekTech, we continue to evaluate new standards like this one in our AI-powered web and enterprise software projects.
