Sightline

Inspiration

Working in cloud consoles like AWS, Azure often turns a simple task into a scavenger hunt across services, settings, documentation, and browser tabs. When I get stuck, I usually leave the console to ask ChatGPT or Claude for instructions. Sometimes the steps are accurate; other times they are outdated, incomplete, or based on a console layout that has already changed. Even with the right answer, I still have to return to the console, locate every resource, and translate those instructions into clicks myself. The AI can describe the maze, but it cannot see where I am or work through it with me. Sightline was inspired by the idea that an agent should operate inside the console’s real context—helping navigate and prepare work while the human remains in control.

What it does

Sightline makes complex operational consoles like AWS, Azure and related, easier to use. It connects every service, resource, and setting involved in a task into one clear path that a person and an AI agent can navigate together, while the human retains control over final changes.

How we built it

We connected ten WebMCP tools directly to Sightline’s shared console state. Agents can search resources, trace dependencies, navigate between services, and prepare coordinated changes while each action appears instantly in the interface for human review. Only the human can apply the final changes.

Challenges we ran into

The hardest part was designing tools that were useful without giving agents too much control. We also had to handle stale revisions, invalid patches, sensitive values, and clear separation between prepared and applied changes.

Accomplishments that we're proud of

• We can now turn a complex operational request into one clear path across multiple services, resources, and settings.
• Agents can find dependencies, explain their impact, and prepare coordinated changes without applying them.
• Humans can review every proposed action inside the console and retain control over the final change.

What we learned

WebMCP works best when tools expose meaningful product actions instead of mirroring buttons. Stable resource IDs, clear schemas, and visible feedback make agent behaviour safer and easier to understand.

What's next for Sightline

Next, we want to connect Sightline to real cloud APIs, authentication, authorization, and audit systems, then test it with larger infrastructure environments and more operational workflows.

Built With

• codex
• react
• typescript
• webmcp