SCRAPR

Inspiration

After seeing the vast number of webscrapers that SerpAPI provides, I realized that this was an opportunity to build a shopping interface, a travel website, even a unique web browser or research platform, without the effort of sourcing the data. However, these ideas I came up with all seemed "leverage AI" to do the same thing... Who wants to see yet another AI agent leverage MCP tools to perform enhanced search? So, instead, I flipped the idea on its head. What if I could make a game where humans race to compete against AI search? Better yet, what if I made a game where humans compete against each other to find obscure facts or shop for the best item? Introducing SCRAPR, where you're the webscraper, looking for obscure facts and shopping for the best products.

What it does

SCRAPR is a multiplayer game that is fully hosted online where users will join a lobby, select between a different settings (Question Difficulty, Re-using words in the prompt in search queries) and 2 different game modes: web search or shopping, and then play a game to find an answer as fast as possible. For search, this means Googling for the answer, trying to get as few searches and clicks and going as fast as possible to get the correct solution. For shopping mode, this means looking for the cheapest and highest quality item that matches the requested description. Be sure to move fast, because once 1 player is done, it starts a timer for everyone else to finish!

The entire game is a visual mock of google chrome in HTML, but remade by hand. It copies the exact details of what you would see in a normal google search.

Users can make their own profiles with a unique username and profile picture. By default, the game lasts for 3 rounds, but this is configurable to be 1, 3, 5, or 7 rounds. Similarly, the timer duration can be set, as well as a few other rules in the game.

Everything works within a single browser tab, without ever needing to leave the game itself. Essentially, I've rebuilt google search and google shopping within a game interface, tracking clicks, searches, word counts, and times, to game-ify the task.

How we built it

The game uses a client–server architecture for multiplayer. One central server (my laptop) hosts every game. The visual "host" of a lobby is just the player who created it, who can change settings and start the game, not a separate machine, but not the real owner. Clients load server-rendered HTML pages with a small amount of vanilla JavaScript. The JavaScript keeps the page current by polling JSON endpoints every 1.5–2 seconds (for example GET /api/games/) and by posting actions (answers, selections, searches, clicks) with a per-player token. Behind a single Flask application (served by gunicorn with one worker and eight threads, and published through a Cloudflare Tunnel) are two layers that run in the same process. The web layer handles every incoming HTTP request. It validates the request, enforces rate limits, checks that the game is actually running, and decides when to call SerpAPI. SerpAPI is used to surface web search, Google Shopping, and autocomplete, with results cached for ten minutes to conserve API credits. The web layer also serves the in-game page viewer, which fetches third-party pages server-side (with a browser-like TLS handshake and a guard against requests to private addresses) and serves a sandboxed copy. It routes game actions and polls to the game layer. The game layer keeps all game state in memory. That means rooms keyed by a five-character code, players, settings, scores, and event logs, plus one game engine per room (quiz or shopping) that defines that mode's rules and scoring. Scores are computed on the server from its own stored copy of the results, so a client can't submit its own prices or ratings. There is no continuously running game loop. Each engine is a state machine, and its timers (round limit, countdown, intermission) advance lazily whenever a request touches the game.

Built With

• flask
• game
• python
• serp