MCP tools

webular mcp has 24 tools. The table marks which ones only read and which ones reach the web.

The tools are grouped by job: Find, Read, Think, Watch and act, Tools and setup, Flow machines, and Low-level HTML. Each tool carries readOnlyHint, plus openWorldHint on a tool that reaches the web. When a tool has no destructiveHint, the table shows “not set”, not “no”.

For agents: llms.txt indexes every page.

The 24 MCP tools, grouped by job
Tool	Use it when	Read-only	Reaches the web	Destructive
Find
`feed`	Read an RSS or Atom feed. Use for a site's newest posts. Returns each item's title, link, and date.	yes	yes	not set
`map`	List a site's URLs from its sitemap and links. Use to see its pages first. Returns the URLs.	yes	yes	not set
`search`	Find pages across the web, Wikipedia, Hacker News, or arXiv. Use for facts that may be new or outside your memory. Returns titles, URLs, and snippets.	yes	yes	not set
Read
`batch`	Read two or more URLs in one call. Use instead of one scrape per URL. Returns one result per URL.	yes	yes	not set
`crawl`	Visit a page and the pages it links to, to a depth and page limit. Use when one page is not enough. Returns each URL and its Markdown.	yes	yes	not set
`media`	Save a page's media to a file: download, screenshot, or PDF. Use for the file, not the text. Returns the path and size.	no	yes	not set
`parse`	Turn a local file, like a PDF or HTML file, into clean text. Use for a file on disk. Returns the text.	yes	no	not set
`scrape`	Read one web page as clean Markdown. Use for one URL; for two or more, use batch. Returns the title and Markdown.	yes	yes	not set
Think
`answer`	Answer a question from live web sources. Use before replying to any question of fact; urls reads only those pages. Returns key sentences and source URLs.	yes	yes	not set
`extract`	Get named parts of a page by CSS selector, like a price, as name:selector pairs or one selector. Use for known layouts. Returns the matched text.	yes	yes	not set
`research`	Research a topic across many sources in one call. Use when asked to research, not search then scrape. Returns key points with source URLs.	yes	yes	not set
`summarize`	Pick the key sentences of a text or page, with no model call. Use for a long text. Returns them verbatim with offsets.	yes	yes	not set
Watch and act
`act`	Read a page in a real browser. Use when a page needs JavaScript. Pass goal instead of url to act in the Cua window named by WEBULAR_CUA_WINDOW. Returns the page text, or status (done or infeasible), at (step not reached), and rungs (ax, ocr, pixel).	no	yes	not set
`monitor`	Check a page for changes since the last check. Use to see what changed. Returns whether and what changed.	yes	yes	not set
Tools and setup
`audit`	Check whether deepsec is installed and show the security scan plan, with no API key. Use for a security scan. Returns readiness and the command to run.	yes	no	not set
`diagram`	List the XState machines that model webular commands. Use to learn how a command flows. Returns names and count.	yes	no	not set
`doctor`	Check that webular can run here. Use when a tool fails for no clear reason. Returns each check's result.	yes	yes	not set
`jev`	Improve webular's Jev calls: learn tests lessons; correct labels one recorded call (id) with the right answer (label). Use only to tune Jev.	yes	yes	not set
`tasks`	List the mise tasks that build, check, and run webular. Use to learn how to run it. Returns task names and count.	yes	no	not set
Flow machines
`machine.next`	Replay events on a machine; return its state and allowed next events.	yes	no	not set
Low-level HTML
`web.fetch`	Low-level: fetch a URL, return raw HTML. To read a page use scrape; use this only to feed web_links, web_markdown, or web_readability.	yes	yes	not set
`web.links`	Extract absolute links from HTML (linkedom).	yes	no	not set
`web.markdown`	Convert HTML to Markdown (Turndown).	yes	no	not set
`web.readability`	Extract main article content (Mozilla Readability + linkedom).	yes	no	not set