TinyFish Drops 30 Open-Source Web Agent Recipes to Kill Playwright Hacks
TinyFish open-sourced a cookbook of 30+ working web agent recipes, from happy hour scrapers to competitor intelligence dashboards, all built on its browser automation APIs.
Takeaways
- TinyFish's open-source cookbook hits 1,900+ stars with 30+ working web agent recipes under MIT license.
- Four endpoints: Search, Fetch, Agent, Browser, all accessible with one API key.
- Search and Fetch are free tier with generous rate limits, no credit card required.
- Recipes span shopping, travel, research, developer tools, education, and finance verticals.
- Includes MCP server, CLI, Python and TypeScript SDKs, plus n8n workflow imports.
- Agents take a URL plus plain-English goal, return structured JSON, replacing hand-rolled Playwright.
Building AI agents that actually navigate the live web, sign into sites, click through forms, and return clean JSON is still a mess of Playwright scripts, rotating proxies, and headless Chrome hacks. TinyFish, a Palo Alto startup building enterprise web agent infrastructure, has been quietly publishing something to short-circuit that grind: a public cookbook repo full of end-to-end sample apps you can clone, tweak, and ship.
The repository has crossed 1,900 stars and now hosts more than 30 standalone projects, each targeting a real workflow: motorbike rental price comparison across Vietnamese cities, semiconductor supply chain tracking, live sports betting odds aggregation, Discord bots for anime figure collectors, and a voice-first academic research copilot that scans ArXiv and PubMed. Every recipe is a working demo, most with a deployed Vercel preview, and the whole thing ships under MIT.
What TinyFish actually is
TinyFish provides enterprise infrastructure for AI web agents, letting you run hundreds of websites simultaneously to navigate, authenticate, extract data, and automate workflows with a fully serverless architecture. The company claims hundreds of thousands of agents run monthly for Google, DoorDash, ClassPass, and Fortune 500 customers.
The platform exposes four endpoints, and the cookbook teaches you when to reach for each one:
- Search at
api.search.tinyfish.ai: structured JSON web search built for LLM consumption, sub-500ms responses. - Fetch at
api.fetch.tinyfish.ai: any URL to clean markdown, JSON, or HTML using a real full-browser render. Failed URLs are not billed. - Agent: pass a URL plus a natural-language goal, the agent navigates and acts, returns structured JSON. Takes 10 seconds to a few minutes.