Single endpoint replaces chaining search, scrape, and LLM APIs
POST /v1/research
Deep research across the whole web.
Multi-source research with citations, fully automated.
webclaw's /v1/research endpoint takes a research question, finds relevant sources, reads them, cross-references facts, and returns a synthesized answer with citations. Perfect for building research agents, competitive intelligence tools, or automated knowledge base updates.
How it works
Build it step by step.
The real flow, one step at a time. Switch between TypeScript, Python, and cURL on any snippet.
- 1
Pose the question
Send a research question to /v1/research, setting depth and the number of sources to read.
// Start an async research job with depth + source countconst job = await webclaw.research.create({ query: "What are the latest developments in Rust async runtimes in 2026?", maxSources: 15, depth: "deep",});console.log(job.id); // poll this id for results - 2
Search and read sources
webclaw runs multi-hop web searches and reads each result, handling bot-protected pages.
- 3
Cross-reference facts
Findings are compared across sources so claims are corroborated rather than taken from a single page.
- 4
Return a cited answer
Poll the async job to get a synthesized summary with source URLs and inline citations.
// Poll the async job until the report is readyconst result = await webclaw.research.wait(job.id);console.log(result.summary); // synthesized answerconsole.log(result.sources); // source URLs with snippetsconsole.log(result.citations); // inline citation references
Why webclaw
Built for deep research.
Multi-hop reasoning across multiple sources
Structured output with citations and source URLs
Async job model for long-running research
Works on bot-protected sites that other tools cannot reach
What you get
Everything this use case needs.
- Multi-hop web search and scraping
- Synthesized answers with citations
- Async job model
- Configurable depth and source count
- Works on bot-protected sources
Where it fits
Built for the messy parts.
Manual research means copying URLs, pasting content into a document, and hoping nothing was missed. Simple search-and-summarize is shallow and misses context across sources. Building a real research agent requires chaining many APIs and handling edge cases.
One endpoint takes a question, plans research, executes multi-hop web searches, reads sources with automatic bot bypass, synthesizes findings, and returns structured output with citations. Async job model means long-running research does not block your application.
Common questions
Frequently asked questions
How does webclaw research compare to a normal LLM with web search?
Standard LLM web search typically reads 1-3 top results and summarizes them. webclaw research performs multi-hop reasoning, follows links, reads 10-30 sources, cross-references facts, and handles bot-protected sites. Output includes full source URLs and citations.
Can I use webclaw research for competitive intelligence?
Yes. Point the research endpoint at questions like 'What features did competitor X ship in Q1 2026?' and it will search, read changelogs, blog posts, and release notes, then return a synthesized timeline with source citations.
How long does a research job take?
Shallow research (5 sources) typically completes in 20-60 seconds. Deep research (15-30 sources) can take 2-5 minutes. Both run as async jobs so your application does not block waiting.
For AI agents
Or hand it to your agent.
Add the webclaw MCP server to Claude, Cursor, or any MCP client, then paste this prompt. The agent calls the webclaw tools and hands the result back to your model — no code to write.
PROMPT FOR YOUR AGENT
Using the webclaw tools, run deep research on [your research question]. Call the research tool to plan the investigation, search the web across multiple sources, read each one (including pages that block ordinary scrapers), and cross-reference the findings so claims are corroborated rather than pulled from a single page; if you need to widen coverage, use the search tool to surface additional sources to feed in. Set the depth to thorough and aim for at least 15 sources. Return a synthesized written answer organized by sub-topic, with inline citations linking each key claim to the source URL it came from, followed by a numbered list of all sources you read. Flag any point where sources disagreed or the evidence was thin so I know what to double-check.
Ready to build? Start extracting.
Cancel anytime. Clean, structured data on every call.