WebMCP, governed

Websites can now expose their own actions as tools AI agents call in the browser. MCP Trail bridges those WebMCP tools into a guarded MCP server—auth, audit, approvals, and drift protection—so browser tools get the same control as everything else.

What is WebMCP?

WebMCP is an experimental browser API (navigator.modelContext, Chrome 150+) that lets a website publish its own actions—like add to cart, change seat, or create invoice—as structured tools an AI agent can call directly in the page, as the logged-in user. It is the web-page sibling of the Model Context Protocol (MCP): MCP standardizes how agents call tools on servers; WebMCP does it for the tools a website exposes to an agent running in the browser.

New to the topic? Start with What is WebMCP?, or, if you build websites, see how to add WebMCP to a website.

The problem: WebMCP tools run with no guardrails

WebMCP tools execute inside the website's page, as the logged-in user, with real side effects. The raw API has no built-in authentication, no audit log, and no approval step. A site can also change its toolset at any time, and there is nothing to stop an agent from running the new tools silently.

That is exactly the gap MCP Trail closes—without any change to the website.

How MCP Trail guards WebMCP tools

The Webmcp Trail Chrome extension bridges a tab's WebMCP tools through the Guardian proxy, turning them into a normal—fully governed—MCP server.

Install the Webmcp Trail extension

Add the Chrome extension (Chrome 150+ with the WebMCP flag) and paste an MCP Trail API key.

Open a site that exposes WebMCP tools

The extension's “This tab” panel detects the tools a website publishes via navigator.modelContext.

Create a guarded server in one click

MCP Trail turns the tab's tools into a browser-bridged MCP server and returns a proxy URL and bearer token.

Run with policy, approvals, and audit

Call the tools from the popup or from Claude / Cursor—every call is governed, logged, and metered.

Add Webmcp Trail to Chrome — free Read the full extension guide in the docs →

What you get, instantly

Once a site is bridged, its WebMCP tools inherit the entire MCP Trail enforcement stack.

Authentication

The bridged server gets a per-server bearer token. The raw browser tools are never exposed directly.

Audit logs

Every tool call is recorded with the tool name, arguments, and outcome—searchable evidence for reviews and incidents.

Human approval (HITL)

Flip a per-tool switch so sensitive browser actions pause in a queue until a person approves them.

Drift attestation

MCP Trail snapshots the approved toolset. If a site silently changes a tool, calls are blocked until you re-approve.

Policy & allowlists

Allow, log, gate, or disable each tool—the same policy engine that governs your MCP servers.

Budgets & rate limits

Every WebMCP call is metered, so browser tools sit under the same cost and abuse controls as everything else.

Go deeper on WebMCP

What is WebMCP? — a plain-English guide to how websites expose tools to AI agents.
How to add WebMCP to a website — register your site's actions with navigator.modelContext, with code.
Turn a WebMCP site into a guarded MCP server — add auth and logs with the extension and MCP Trail.
Human-in-the-loop approvals for WebMCP — pause sensitive browser-tool calls until a person approves.

Bring your browser tools under control

Bridge any WebMCP site into a guarded MCP server—authentication, audit, approvals, and drift protection, in a few clicks.

Open MCP Trail Explore Features