Set up WebMCP

WebMCP lets a website expose structured, permissioned tools to compatible browser agents. CrawlConsole helps you configure those tools and measure registrations, calls, failures, and active sessions.

Check readiness first

Use the AI Agent Readiness Scanner to inspect a public site before installing WebMCP. The scan checks whether the site exposes the expected signals and returns a public result page that can be shared with a developer.

Open the WebMCP workspace

1. Open a CrawlConsole property.
2. Expand WebMCP in the property navigation.
3. Create or select a configuration.
4. Add tools with clear names, descriptions, and input contracts.
5. Add path rules that limit where each tool can be available.

Tool descriptions should explain when an agent should call the tool, what it returns, and any preconditions. Avoid broad descriptions such as "manage the website."

Design safe tools

Start with read-only actions that have a narrow effect. Good first tools include:

• retrieve a product or article by canonical id;
• search public catalog content;
• check inventory or availability;
• inspect a page-specific structured record;
• submit a low-risk support or lead form with explicit confirmation.

Require authentication and a clear user confirmation before tools create orders, change account data, send messages, or perform other consequential actions.

Apply path rules

Path rules determine which pages can register a tool. Use the smallest pattern that covers the intended experience.

/products/*
/collections/*
/support/contact

Do not register every tool on every page by default. Page-specific registration gives agents better context and reduces accidental tool selection.

Monitor the installation

WebMCP analytics separates:

• tag loads;
• tool registrations;
• tool calls;
• successful and failed calls;
• active sessions;
• recent failure messages.

Use get_webmcp_analytics through MCP when an agent needs the same property-level view.