[ free tool ]
Find the answer on a page
Enter a page and a question. Instead of the whole page, you get the few passages that answer it, ranked, each with the section it came from, the same way an agent reads with Clawifi.
What it does#
- Reads the page through Clawifi, splits its main text into passages and ranks them against your question.
- Shows up to 3 passages, best match first, each labelled with its rank and the section heading it sits under, and says how much of the page was kept.
- Shows the passages as text, and as Markdown in the same order you can copy or download, ready to quote with the page’s address. When no passage shares a word with the question, it says so instead of showing the page.
Limits#
- The ranking matches the words of your question against the page; it does not write an answer, and a passage can rank well without answering the question. Read them before you quote them.
- Up to 3 passages in the free tool. With a key,
top_ksets how many. - Public pages only, a question of up to 300 characters, and 12 runs a minute per address shared with the other free tools.
Do this from your agent#
This is the query parameter of scrape_url. It works on the hosted MCP server without a key as a trial (up to 3 passages, 12 a minute and 50 a day per address); with a key, set top_k.
scrape_url(url="https://en.wikipedia.org/wiki/Model_Context_Protocol", query="When was MCP announced, and by whom?")curl https://api.clawifi.dev/scrape \
-H "Authorization: Bearer $CLAWIFI_API_KEY" \
-H "Content-Type: application/json" \
-d '{"url":"https://en.wikipedia.org/wiki/Model_Context_Protocol","formats":["markdown"],"query":"When was MCP announced, and by whom?","top_k":5}'from clawifi import Clawifi
client = Clawifi() # reads CLAWIFI_API_KEY
page = client.scrape(
"https://en.wikipedia.org/wiki/Model_Context_Protocol",
formats=["markdown"],
query="When was MCP announced, and by whom?",
top_k=5,
)
print(page["formats"]["markdown"])Read more: Fetch and scrape · POST /scrape · Remote MCP · Docs · Pricing
Questions#
Why return passages instead of the whole page?
A long page can be tens of thousands of characters, most of it unrelated to the question. Sending only the passages that answer it keeps an agent’s context small and makes the source of each claim easy to cite.
Does it write an answer for me?
No. It finds and ranks the passages of the page; it does not generate text. Your agent or model writes the answer from them.
How many passages can I get?
The free tool and the keyless MCP trial return up to 3. With an API key, the top_k parameter of scrape_url and POST /scrape sets the number.
What if no passage matches?
The result says that no passage on the page answers the question, and nothing is shown as a passage. Try the words the page itself is likely to use, or read the whole page with URL to Markdown.