DocsOpen MCP
Open MCP tools
Your agent's own model reads each page and decides every step; pilotfish carries out the action, checks it, and reports what changed.
Read the page
browser_navigate- Open a URL. Returns what happened and the visible screen as a compact snapshot.
browser_snapshot- The page as a compact tree with refs, around the viewport or the whole page.
browser_find- Refs of elements on the whole page that match words, best first.
browser_query- Read values with a read-only JavaScript expression: the cheapest way to read.
browser_get_text- Readable text of the page (PDFs too), with secrets masked.
browser_screenshot- A JPEG of the viewport (off when secrets are configured).
Act on it
browser_click- Click by ref. Refuses rather than clicking the wrong thing (covered, stale, hidden).
browser_type- Replace a field's text; {{name}} placeholders type configured secrets.
browser_select- Pick options in a <select> by label or value.
browser_press- A key or shortcut: Enter, Escape, Control+a.
browser_scroll- Scroll about one screen; returns what it shows.
browser_actions- Up to 10 steps in one call, each checked like the single tools.
Wait and recover
browser_wait- Wait 0.5 to 10 s, or until a text appears or goes, then return what changed.
browser_resume- After NEEDS A PERSON: waits until the page has moved on, then returns it.
browser_tabs- List open tabs, or switch to another.
browser_close- Close the browser; the next call starts a fresh one.