The Chrome for Developers team has published guidance on annotating WebMCP tools so that browser agents can make better UI decisions while better protecting users.
Chrome for Developers · WebMCP
Annotating WebMCP tools to guide safer browser agents
Chrome's developer team says metadata attached to exposed tools can help browser agents pick the right interface elements and reduce the risk of unsafe actions taken on a user's behalf.
How tool annotations are meant to work
Step 1
Developer annotates tools
Metadata is attached to each tool exposed to browser agents.
→
Step 2
Agent chooses elements
Annotations help the agent select the right interface element for the task.
→
Step 3
Fewer unsafe actions
Clear metadata is intended to reduce the risk of unintended actions for the user.
What the guidance offers
A concrete place for developers to act on agent safety
Better UI decisions by browser agents when completing tasks
A safeguard as agents move from experiments to everyday use
What remains unverified
No independent reviews or third-party testing yet
Detailed specification of annotation fields not available
How agents interpret each annotation is not described
Developer adoption levels are undocumented
Key takeaway
Whether annotations become a convention depends on real-world agent behavior and on broader standards work.
Agent behavior
How browser agents use annotations in practice
Standards
Whether the guidance is backed by wider standards work
Platform role
Agent safety handled in web platform tooling, not only by model providers
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...