connect/tinyfish
TinyFish search, extraction and asynchronous browser-agent tools for berm services
Tools
Readme
TinyFish
Search the web, news and academic publications, extract pages, or run an asynchronous browser agent using the deployment’s TinyFish account.
Call
With Crab signed in to Cloud, replace owner/tinyfish with the address of a running deployment you can access. If this page only offers a template, deploy a copy first.
crab call owner/tinyfish search '{"query":"Rust WebAssembly hosting"}'
Returns a page of search results. Use search_news or search_publications for those sources, fetch to extract URLs, and run for browser tasks.
On a callable deployment, you can also open a tool on this page to call it, or connect your agent through CrabTalk MCP. Calls use this deployment’s provider credentials.
Deploy your own
If this page offers Deploy your own, use it to create a private copy with your own credentials and state. The CLI equivalent is:
crab deploy --from owner/tinyfish
Set TINYFISH_API_KEY through credential setup. Access and usage charges depend on the connected TinyFish account and wallet.
Template deployment needs a published image. To build a modified version or deploy when no template is published, use the source instructions.
Usage notes
fetch accepts 1–10 URLs and returns Markdown, HTML or JSON. Inspect both results and errors for partial failures. Conditional fetches accept one URL and may return null content when unchanged.
run returns a run ID. Poll run_status explicitly or call run_cancel. Enable screenshots or HTML in capture_config before calling step_screenshot or step_html. Screenshots return base64 JPEG; HTML returns text.
Failed or cancelled runs may still incur provider charges for completed steps. Calls make one upstream request, with explicit pagination and polling and no automatic retries.
Dedicated batch, monitor, wallet, profile, vault, synchronous automation, SSE, research-report, dataset and browser-session APIs are not included.