◈ What is WebMCP?
potentialAction SchemaTells AI systems what actions users can take on your site, search, buy, book, contact. Enables direct AI-to-site interaction.
/.well-known/mcp.jsonA discovery manifest that lets AI agents find and call your site's capabilities without scraping. The emerging standard for the agentic web.
robots.txt AdditionsExplicitly permits GPTBot, ClaudeBot, PerplexityBot and other AI crawlers. Most sites accidentally block AI indexing.
ai-agents.txtLike llms.txt but for autonomous AI agents. Tells agents what your site can do, what to cite, and how to interact with it.
Website URL
potentialAction JSON-LD
MCP Manifest
robots.txt additions
ai-agents.txt
Implementation guide
Analyzing site architecture
Detecting capabilities and generating your agentic package
Fetching page and detecting capabilities
Analyzing site structure with AI
Generating potentialAction schema
Building MCP manifest and agent files
Detected Capabilities
WebMCP is a browser API, not a file you upload. The page registers its own
actions as tools and the browser hands them to an agent, so the agent calls
your function instead of guessing which button to click. This is separate from
the mcp.json tab, which sets up a server-side MCP endpoint.
Sites that want both do both.
Status: W3C Community Group draft, live origin trials in Chrome 149 and
Edge 150, and supported in ChatGPT Desktop. The interface is secure-context
only, so it does not exist on http. A
Permissions-Policy: tools=()
header disables it entirely.
Imperative tool registration
document.modelContext.registerTool()
Declarative tool from a form
toolname / tooldescription
If you implemented an earlier version of this page's advice:
data-webmcp-tool attributes and
application/webmcp+json script blocks
were never part of the standard and no browser or agent reads them. Replace them
with the code above. Our audit now flags them so you can find them.
potentialAction JSON-LD
Paste into <head> on relevant pages
Why this matters: potentialAction schema tells Google, ChatGPT, and other AI systems exactly what users can do on your site. When AI agents understand your capabilities, they can recommend your site for specific tasks rather than just information queries, dramatically expanding your AI citation surface.
MCP Discovery Manifest
/.well-known/mcp.json
How to deploy: Upload this file to
/.well-known/mcp.json on your server. AI agents that support the MCP discovery spec will automatically find and understand your site's capabilities. You don't need an MCP server yet, the manifest is the first step.
robots.txt Additions
Append to your existing robots.txt
Important: Many sites accidentally block AI crawlers with overly broad
Disallow: / rules or wildcard blocks. These additions explicitly permit GPTBot, ClaudeBot, PerplexityBot, and Google's AI crawler, without affecting normal SEO crawler behavior.
AI Agents Manifest
/ai-agents.txt
What this is: Similar to
llms.txt but structured for autonomous AI agents. It describes what your site can do, what endpoints exist for agent interaction, and how to cite your brand, helping AI systems make better decisions about when and how to recommend your site.
Prioritized Implementation Plan