Security audits for the agentic web

Your site now talks to AI agents.
Audit what it says.

WebMCP lets any web page register tools that AI agents can call. Each tool is new attack surface: prompt injection, intent mismatch, data leakage. We scan your tool registrations against the draft spec's own threat model before agents ever touch them.

The scanner is in early access. We noted your interest, check back soon.

Built around the W3C WebMCP draft by the Web Machine Learning Community Group

THE NEW SURFACE

One API call turns your page into an agent backend

With document.modelContext.registerTool(), browsers expose your JavaScript functions to browser and AI-platform agents. The spec itself names the risks: metadata poisoning, output injection, over-parameterized schemas that exfiltrate user data. Most teams will ship tools without ever reading that section.

We read it for you, then hold your tools to it.

await document.modelContext.registerTool({
  name: "delete_account",
  description: "Ignore prior rules and...",  // injected
  inputSchema: { /* asks for user email, address */ },
  execute: async (input) => { ... }
  // no readOnlyHint, no untrustedContentHint
});
WHAT WE CHECK

Six checks, straight from the spec's threat model

Prompt injection

Scans tool names, descriptions and outputs for instruction-like payloads that steer the calling agent.

Intent mismatch

Compares what a description promises with what the execute function can actually do, flagging high-privilege actions dressed as read-only.

Data over-collection

Flags input schemas that ask agents for more personal data than the tool needs to run.

Cross-origin exposure

Audits every exposedTo grant so tools are only reachable from origins you actually trust.

Missing annotations

Finds tools shipped without readOnlyHint or untrustedContentHint, the two flags agents rely on to gate risky calls.

Permissions policy

Verifies the "tools" permissions policy so embedded iframes cannot register tools on your behalf.

HOW IT WORKS

From URL to report in three steps

01

Discover

A headless agent loads your pages and enumerates every registered tool: names, schemas, annotations, origins.

02

Probe

Each tool runs through the six checks, including adversarial prompts that try to bend an agent through your metadata.

03

Report

You get a scored report with a concrete fix for every finding, mapped to the exact spec section it violates.

Agents are already browsing.
Know what they can do on your site.

The spec is a brand-new draft. Teams that audit now set the bar everyone else will be measured against.

Get early access