Submission for the Google Cloud Rapid Agent Hackathon — "Building Agents for Real-World Challenges". Partner track: MongoDB MCP.
Building this in the open — join the team chat on Discord: https://discord.gg/RTBbx3DjqC
Picking the right robotic lawn mower for a specific yard — and getting it set up to actually work there — is mostly trial and error today. Buyers and installers juggle yard size, slope, obstacle layout, boundary type, charging access and model-by-model spec sheets, then guess at install steps. Most of that knowledge is locked in product PDFs, forum threads and people's heads.
An agent that takes a description of a target yard and returns:
- A short list of suitable mower models from a curated registry, with the reasons each one fits (yard area, slope tolerance, obstacle handling, boundary technology, charging needs).
- A deployment plan for the chosen model — boundary placement, charging dock location, first-mow zones, expected schedule.
- A persistent record of the recommendation written back to the registry so later jobs can learn from past deployments.
The registry of mower models, yards, and past deployment plans lives in MongoDB and is exposed to the agent through the MongoDB MCP server.
- Model: Gemini (Google Cloud)
- Orchestration: Google Cloud Agent Builder
- Data + MCP: MongoDB collections —
mower_models,yards,deployment_plans— via the MongoDB MCP server
- Hosted project URL
- Public repository
- LICENSE detectable at the top of the repo — Apache-2.0
- ~3 minute demo video
- Selected partner track — MongoDB MCP
- Completed Devpost submission form
🚧 Early development. Issues, milestones, and design notes are tracked in this repo's Issues tab.
Contributions welcome — see CONTRIBUTING.md. Day-to-day discussion happens on the project Discord.