SNIPR WebMCP

Inspiration

Renovation platforms contain a lot of operational knowledge, but that knowledge is often locked inside dashboards, planning tools, room editors, documents, and product choices. We wanted to show what happens when an AI agent can use WebMCP to access that context safely and directly.

What it does

SNIPR WebMCP lets an AI assistant answer real questions about a renovation assignment. In the demo, a contractor asks which shower mixer belongs to a specific bathroom. WebMCP reads the selected SNIPR assignment, address, room data, and resident choices, then returns the exact product information.

The demo also shows the answer visually: the user opens the execution map, selects the address, opens the 3D building view, moves to top view, selects the bathroom, and clicks the shower element to see the attached specifications.

Finally, the project demonstrates safe actions. WebMCP can propose an operational action, such as finding an inspection slot, but SNIPR requires explicit human approval before anything is written back.

How we built it

We integrated WebMCP endpoints into SNIPR so agents can access structured assignment context, address data, room information, planning data, communication state, documents, catalog information, and controlled actions.

The frontend demo combines the normal SNIPR dashboard, WebMCP chat, an execution map, and a 3D building/room view. The video script uses Playwright and FFmpeg to record a reproducible demo with synchronized scene-based voice-over.

Challenges we ran into

The hardest part was making the demo coherent. Browser loading, 3D interactions, voice-over timing, and human approval flows all needed to line up. We also had to make sure WebMCP felt like part of the real operational workflow, not just a chatbot bolted onto the side.

Accomplishments that we're proud of

We are proud that the demo shows three important things together: structured context, visual proof, and controlled writes. The agent can answer useful questions, the user can verify the answer spatially, and the system still protects operational changes with human approval.

What we learned

WebMCP becomes most valuable when it connects AI to the existing source of truth, instead of replacing it. The best experience is not just “AI answers a question”, but “AI helps the user navigate, verify, and act inside the real system.”

What's next

Next steps are deeper WebMCP coverage, more robust permission scopes, richer 3D evidence, and production-grade audit trails for every agent-assisted action.

Built With

• ffmpeg
• maplibre
• node.js
• nuxt
• openstreetmap
• playwright
• tailwind
• three.js
• typescript
• vue
• webmcp