# WebMCP Embeds (Beta)
:::caution[Beta]
WebMCP embeds are an early-stage feature. The underlying [WebMCP proposal](https://github.com/webmachinelearning/webmcp) shipped in Chrome 146 in February 2026 and the consumer ecosystem (Chrome's built-in AI agent, browser extensions like the [Model Context Tool Inspector](https://github.com/beaufortfrancois/model-context-tool-inspector)) is still maturing. Expect rough edges and some manual flag-flipping in the browser. The producer-side API (what Wire ships) is stable; the consumer-side experience varies by browser and agent.
:::
WebMCP embeds let you drop a single `
```
That's the whole integration. On page load the shim:
1. Reads `data-key` and validates the format.
2. Feature-detects via `'modelContext' in navigator`. If absent, logs a hint and exits.
3. Fetches `/v1/embed/tools` from the Wire API to load the schemas your key is scoped to.
4. Calls `navigator.modelContext.registerTool` for each one, with a single `AbortController` signal so all tools can be unregistered together.
## Verify it works
The fastest way to confirm your embed is registering tools is the [Model Context Tool Inspector](https://chromewebstore.google.com/detail/model-context-tool-inspec/gbpdfapgefenggkahomfgkhfehlcenpd) Chrome extension by François Beaufort. It opens a side panel that lists every tool the current page has registered, lets you inspect schemas, and lets you invoke any tool with a JSON args payload.
Setup:
1. Install the extension from the Chrome Web Store.
2. In `chrome://flags`, enable both **WebMCP** and **WebMCP for testing** (the Inspector reads via `navigator.modelContextTesting`, gated behind that flag).
3. Restart Chrome.
4. Open the page where you embedded Wire's `