Hey HN, I'm Arjun, one of the founders of rtrvr.ai.

For the past year, we've been building and iterating on an in-browser AI agent. But a few months ago, we hit a wall in our approach and decided to tear everything down and rebuild from first principles. The result is what we're re-launching today.

The Problem/Insight: We kept coming back to a simple, nagging question: web agents by themselves are a nice to have but we kept getting requests on possible ways to integrate or call our agent from another agent. Users also kept asking to reuse their own existing chatbot subscriptions to power the agent.

The "Aha!" Moment with MCP: This led us down the rabbit hole of interoperability and we landed on MCP. Our core insight was this: what if we exposed the Chrome Extension itself as a Remote MCP server?

This simple idea completely changed our architecture and value proposition. It turns the user's browser into a controllable, sandboxed endpoint that any other AI can talk to with a simple copy/paste of a MCP url.

What this new architecture unlocks: Remote Control Your Browser: You can stay in your flow in Claude and tell it to perform tasks on your behalf in background tabs. No more context switching. Tell Claude to file that Jira ticket or scrape a site without ever leaving the chat.

Reuse Your Own Subscription (The "BYO-Sub" model): Since Claude is driving the action, you're using your existing subscription to power the agent for basic tasks. We see this as the future – our agent provides the secure execution rails, and you bring the brain you already pay for. Users can then upgrade to our more advanced, multi-step agent for complex workflows when needed.

Our Bigger Bet - The Agentic Web: We believe the future isn't a single "god agent" but a network of specialized agents that collaborate. By exposing the browser as a common execution layer, we're hoping to provide the "rails" for this future. We're already in talks with other agent platforms to build on this, enabling a more open and interconnected ecosystem.

We just launched on Product Hunt today and would genuinely love to hear the HN community's feedback. What are the security implications we haven't thought of? What are the unlocked use cases you see by giving an agent/app browser context? What's the next logical step for an open "Agentic Web"? All criticism and ideas are welcome.

Thanks for reading!

Links: Product Hunt Launch: https://www.producthunt.com/products/rtrvr-ai?launch=rtrvr-a...

Chrome Extension: https://chromewebstore.google.com/detail/rtrvrai-ai-web-agen...