Outline - Web MCP And The Agentic Web

Speaker

Maximiliano Firtman discusses Web MCP and agent interaction with web applications.

Safety Status

This is a safety-redacted public outline. Runnable examples and browser bridge mechanics are omitted.

Thesis

Agents need explicit web-facing contracts so they can understand and interact with applications more reliably than through visual inference alone.

Section Map

1. The agentic web: agents build for, run inside, and browse the web.
2. Current web interaction methods are costly and brittle.
3. Moving from inference to explicit contracts.
4. Difference between backend integrations and frontend-declared capabilities.
5. Tool-like page capabilities at a conceptual level.
6. Imperative and declarative approaches discussed conceptually.
7. Demo summaries with operational mechanics removed.
8. Design guidance for small, focused capabilities.
9. Experimental status and safe adoption path.
10. Wrap-up: faster and cheaper agent access requires careful exposure.

Concepts

• Agentic web: Web applications and agents increasingly interact in both directions.
• Explicit contract: A page offers a narrow capability rather than forcing the agent to infer everything.
• Frontend-declared capability: A capability tied to the current page state rather than only a backend API.
• Read-only diagnostic start: Begin with observation before allowing state changes.
• Small output: Return only what the agent needs for the task.
• State-aware availability: Expose capabilities only when relevant.

Not Included

• Runnable code.
• Exact API object shapes.
• Browser bridge setup.
• Demo commands.
• State-changing examples from the source talk.

Safe Use

Use this talk to plan safe adoption, evaluate privacy impact, and design narrow page capabilities at a conceptual level.