Inspiration
Finding new games to play is hard. Watching youtube lists, twitch streams or reading game blogs. many times the games you like may be out of your budget or may not even work on your PC. Hence the idea to create a AI agent that searches for your games according to your specifications is very much needed.
What it does
It uses two MCP servers, the official MongoDB MCP to interact with a steam games database, and another unoffical steam-mcp developed by TMHSDigital
How we built it
I intially looked for a steam database for games on kaggle but seeing that the database missed the requirements field, we enriched the original database with a enrichment script that looks for requirements from the steam website. Because the free tier of MongoDB only has a limit of 512 MB we cut down the list of 122,000 games to 9127 games.
Challenges we ran into
The biggest issue is deployment. Idk why but i just cant seem to get the mcp tools to work on google cloud run. i tried out some solutions and nothing seems to work. im uploading the bare bones agent but my video does cover the local running which uses the tools i mentioned above. another issue was Latency! vertex ai was sluggish with queries first, which made testing very difficult at first. i also had issues with deployment and creating a frontend. both of which did not turn out very well. Also the semantic searching for mongoDB gave me a bit of an issue at start as i was unfamiliar with how it works.
Accomplishments that we're proud of
Super happy to create my first agent using google-adk! im looking forward to making more of these in the future! with more features and compex tasks achievable!.
What we learned
Learned a lot about semantic searching, database indexing, how ai agents worked, subagents, orchestration, cloud deployment, scale, latency and so much more!
What's next for game_scout
A proper frontend, fixing the deployment and hopefully adding more features.
Built With
- google-adk
- google-cloud
- mcp
- mongodb
- python
Log in or sign up for Devpost to join the conversation.