Agno's AnakinTools sends scrape, crawl and map requests to Anakin.io as jobs, then polls each job every poll_interval seconds (2 by default) until it completes, so the model gets page content or links back from one tool call. Search is a single synchronous request.
| Method | Flag | Anakin endpoint |
|---|---|---|
scrape_url | enable_scrape | POST /url-scraper |
crawl_website | enable_crawl | POST /crawl |
map_website | enable_map | POST /map |
search_web | enable_search | POST /search |
Only scrape_url is on by default, and all=True turns on all four.
from agno.agent import Agent
from agno.tools.anakin import AnakinTools
agent = Agent(tools=[AnakinTools(all=True)], markdown=True)
agent.print_response("Scrape https://docs.agno.com/introduction/ and summarize it")
agent.print_response("List the URLs reachable from https://docs.agno.com")Against a stand-in for the Anakin API, one scrape_url call submitted the job and checked it three times before handing the result back.
# real output: AnakinTools(poll_interval=0.01).scrape_url("https://docs.agno.com/introduction")
POST /url-scraper {'url': 'https://docs.agno.com/introduction', 'generateJson': False, 'useBrowser': False, 'country': 'us'}
GET /url-scraper/job_1
GET /url-scraper/job_1
GET /url-scraper/job_1
scrape_url -> {"status": "completed", "markdown": "# Agno\nBuild agents..."}max_wait_time (120 seconds by default) caps the wait. A job that fails or runs past it comes back as an error string the model can read.
# real output: AnakinTools(poll_interval=0.01, max_wait_time=0.05) on a job that never finishes
Error scraping https://example.com: Job job_2 was still 'processing' after 0.05s. Retry later or raise max_wait_time.Scrape and crawl route through a proxy in the country you set ("us" by default), and use_browser=True renders JavaScript-heavy pages in a headless browser.
See the cookbook, and learn more about toolkits in the documentation.
Frequently asked questions
Set ANAKIN_API_KEY, import AnakinTools from agno.tools.anakin and pass tools=[AnakinTools()] to your Agent. It uses httpx, which Agno already depends on, so no vendor SDK is needed.
Only scrape_url. Turn on crawl_website, map_website and search_web with enable_crawl, enable_map and enable_search, or pass all=True for all four.
Nothing until the job settles. The toolkit polls the job and returns the finished JSON. A job that fails, or is still running after max_wait_time seconds (120 by default), returns an error string in place of the result.
Shipped around the same time
October 12, 2026
Changelog
v3.1.3ClaudeAgent and CodexAgent take typed native options
Ashpreet Bedi, Kaustubh Shukla, and 1 other
1 min read
October 12, 2026
Changelog
v3.1.3Claude and Codex sessions take one turn at a time
Kaustubh Shukla, Ashpreet Bedi, and 2 others
2 min read
October 12, 2026
Changelog
v3.1.3Claude subagent tool calls point to the tool that delegated them
Kaustubh Shukla, Yash Pratap Solanky
1 min read