Why a New Web Tool for AI Agents Is Raising Red Flags for SEO

Search Engine Journal recently reported that a new protocol called WebMCP—designed to let websites hand AI agents ready-made tools—can actually be used to hijack those agents. The security advice from Chrome warns that any website exposing these tools must take responsibility for keeping them safe, because the AI models inside the agents cannot reliably tell the difference between real data and hidden commands.

This matters to anyone who cares about search rankings and AI SEO. As Google and other search engines start using AI agents to browse websites, how you structure your site for those agents directly affects how your business appears in results. The same tools that make your site easier for an AI to understand can also become an open door for attackers to manipulate what the agent sees and does.

The Real Risk: Your Own Website Content Can Be Used Against You

Here is the issue that most businesses will actually face. The attack does not need a malicious website. It works through your own user-generated content—product reviews, comments, forum posts, support replies. If someone plants an instruction inside a review, your legitimate tool happily passes that instruction to the AI agent. The agent treats it as a command, not just text.

This is not a bug that can be patched inside the AI model. Chrome’s guidance makes clear that the probabilistic nature of large language models means they cannot reliably separate data from instructions. So the responsibility falls entirely on the website that exposes the tools. For Australian businesses trying to rank well on Google while also adopting AI-friendly features, this creates a new layer of risk that must be managed alongside traditional SEO work.

What This Means for Australian SMBs

If you run a small or mid-sized business in Australia, you might be tempted to make your website “agent-ready” by adding WebMCP tools that help AI assistants find your product info, prices, or bookings. That could give you an edge in AI-powered search results. But without proper security, you could accidentally turn your own customer reviews or blog comments into a weapon that misdirects the agent—or worse, steals data.

Many SMBs rely on third-party platforms for reviews, comments, or support forums. Those platforms are not necessarily secure either. An attacker can inject a hidden command into a review on any site, and if your tool returns that review, the agent follows it. The result could be a ruined reputation or lost sales, and your Google rankings could suffer as the agent learns your site is unreliable.

What You Can Do Now

  • Audit every tool you expose via WebMCP. Ask yourself: could this return user-generated content, external data, or any text I did not carefully write? If yes, annotate it with the "untrustedContentHint" flag.
  • Mark tools that only read data (not change it) with "readOnlyHint". That tells the agent to ask for user confirmation before acting on the tool’s output.
  • Restrict which AI agents can use each tool by setting the "exposedTo" field to only trusted origins—for example, your own domain or verified partner sites.
  • Cap your tool descriptions at 500 characters and outputs at roughly 1,500 characters, as Chrome recommends. That limits the room for hidden commands.
  • Talk to your web developer or SEO agency about threat-modelling each tool like you would a public API endpoint. Do not ship a tool until you have answered: what untrusted content could this return, and have you marked it?

Getting your site ready for AI agents is a smart move for modern SEO, and tools like WebMCP can give you a real advantage. But the security work is yours to do. MS&VG’s AI-powered SEO service can help Australian SMBs navigate this new landscape safely, so you get the ranking benefits without exposing your business to unnecessary risk.