Skip to main content

Overview

TinyFish tools provide web infrastructure for AI agents: structured web search, full-browser page fetching, and a natural-language browser agent that navigates real websites and extracts structured data.

Key Features

  • TINYFISH_SEARCH
    • Search the web with a query string and get structured JSON results (title, URL, description, snippet)
    • Control the number of results (num_results, 1–50, default 10) and geo-target results with a two-letter country code (country, e.g. us, jp)
  • TINYFISH_FETCH
    • Render and extract clean content from one or more URLs using a full browser — works on JavaScript-heavy pages
    • Pass a single URL string or a JSON array of URLs (urls), and choose the output format (format: markdown (default), html, json, or text)
  • TINYFISH_AGENT
    • Run a browser automation agent: give it a starting URL (url) and a natural-language goal (goal), and it navigates the site and extracts structured data
    • Configure the maximum run time with timeout_seconds (default 300)
    • Returns the stream of agent events plus the final result

Authentication

All TinyFish tools authenticate with a TinyFish API key, set via the api_key config (stored as a secret). If the config is not provided, the tools fall back to the TINYFISH_API_KEY environment variable. Note: Treat API keys as sensitive information and never commit them to public repositories.

Usage Examples

Example: Search the web and fetch the top pages

Example: Extract structured data with the browser agent

Notes

  • TINYFISH_AGENT is an at-least-once operation: if a worker dies mid-run, the runner may resubmit the job. Avoid using it for irreversible side effects (form submissions, purchases, etc.) without an external idempotency guard.
  • TINYFISH_FETCH accepts urls as a single URL string, a comma-separated list, or a JSON array.
  • TINYFISH_SEARCH returns at most 50 results per call.