MCP tools
A hub exposes these 14 tools. Some, such as search_collection and agent, need hub storage or an LLM key you configure.
| Tool | What it does |
|---|---|
scrape |
Fetch one URL and return clean Markdown (or JSON), rendering JavaScript when needed |
fetch_url |
Fetch one URL with fewer transforms: status, headers and body |
fetch_batch |
Fetch many URLs in one call |
map |
List a site’s URLs from its sitemaps and links, optionally filtered |
crawl |
Follow links from a start page within limits you set |
extract |
Pull structured fields from a page into JSON |
search |
Web search through your own Serper, Brave or Tavily keys, or SearXNG for $0 |
search_collection |
Full-text search over pages you’ve already collected |
jobs |
Search ATS job boards and return postings with normalised pay ranges |
interact |
Drive a page in the browser (click, type, scroll) and return what it shows |
read_artifact |
Read a stored large result (PDF text, crawl output) in chunks |
find_extractors |
Show which site adapter would handle a URL |
agent |
Multi-step research with your own LLM (OpenAI, Anthropic, Gemini, Ollama and others) |
list_agents |
Show connected worker nodes and what each can do |
Every result reports whether the page was blocked (by robots.txt or a challenge), so your
agent can tell “no content” apart from “not allowed”.
For parameters and examples, see docs/USAGE.md.