Extraction

snapshot

Compact, agent-friendly view of the page: interactive elements (links, buttons, inputs, selects, custom roles) and headings, each with a stable ref that every other tool accepts as selector (ref:e12). Far cheaper than get_html for deciding what to do next.

Parameters:

Name	Type	Required	Description
mode	string	No	interactive (default) or full (adds images, labels, paragraphs, alerts)
selector	string	No	Restrict the snapshot to a subtree
include_hidden	boolean	No	Include hidden elements (default: false)
max_elements	integer	No	Maximum number of elements (default: 200)
format	string	No	text (default) or json
session_id	string	Yes	Session ID to use

Example Request:

{ "name": "snapshot", "arguments": { "session_id": "uuid-1234" } }

Example Response (text):

{
  "url": "https://example.com/signup",
  "title": "Sign up",
  "count": 6,
  "total": 6,
  "truncated": false,
  "snapshot": "[e1] heading(1) \"Create your account\"\n[e2] textbox \"Email address\" type=email placeholder=\"you@example.com\"\n[e3] checkbox \"newsletter\" checked\n[e4] combobox \"plan\"\n[e5] button \"Create account\"\n[e6] link \"Documentation\" href=/docs"
}

Example Response (json): elements is an array of { ref, role, tag, name, selector, href?, type?, placeholder?, value?, checked?, disabled?, level? }.

Notes:

• When a CAPTCHA widget is on the page, the response carries "captcha": ["turnstile"] (also recaptcha, hcaptcha) and the text snapshot starts with CAPTCHA detected: turnstile (solve_captcha)
• Refs are stored on the elements (data-fmcp-ref) and stay stable across snapshots until the page navigates
• Typical loop: snapshot → click ref:e5 → wait_for_text → snapshot
• Names follow accessibility rules: aria-label, <label for>, placeholder, name, then visible text
• Radio buttons also show their value, so unlabeled buttons of one group stay distinguishable
• Password inputs retain their name, type and ref, but their value is omitted in both text and JSON snapshots

get_text

Extract text content from one or more elements.

Parameters:

Name	Type	Required	Description
selector	string	Yes	CSS selector, XPath (xpath: prefix) or snapshot ref
multiple	boolean	No	Extract from all matching elements (default: false)
wait	number	No	Seconds to wait for the element (default: 5)
raw	boolean	No	Return the text exactly as in the DOM (default: false, whitespace is collapsed and trimmed)
session_id	string	Yes	Session ID to use

Example Request (Single Element):

{
  "name": "get_text",
  "arguments": {
    "selector": "h1.title",
    "session_id": "uuid-1234"
  }
}

Example Response (Single):

{
  "text": "Welcome to Example"
}

Example Request (Multiple Elements):

{
  "name": "get_text",
  "arguments": {
    "selector": "li.item",
    "multiple": true,
    "session_id": "uuid-1234"
  }
}

Example Response (Multiple):

{
  "texts": [
    "First item",
    "Second item",
    "Third item"
  ],
  "count": 3
}

Notes:

• Supports both CSS selectors and XPath
• Returns array when multiple: true
• Throws error if no elements found

get_html

Get HTML content of the page or a specific element.

Parameters:

Name	Type	Required	Description
selector	string	No	Selector of the element to get the outerHTML of
max_length	integer	No	Truncate the HTML to this many characters (response has truncated: true)
session_id	string	Yes	Session ID to use

Example Request (Full Page):

{
  "name": "get_html",
  "arguments": {
    "session_id": "uuid-1234"
  }
}

Example Response (Full Page):

{
  "html": "<!DOCTYPE html><html>...</html>",
  "url": "https://example.com"
}

Example Request (Specific Element):

{
  "name": "get_html",
  "arguments": {
    "selector": "div.content",
    "session_id": "uuid-1234"
  }
}

Example Response (Element):

{
  "html": "<div class=\"content\">...</div>",
  "selector": "div.content"
}

Notes:

• Omit selector to get full page HTML
• Returns outerHTML for specific elements (includes the element itself)

screenshot

Take a screenshot of the page or a specific element.

Parameters:

Name	Type	Required	Description
selector	string	No	CSS selector to screenshot specific element
full_page	boolean	No	Capture full scrollable page (default: false)
format	string	No	Image format: png or jpeg (default: png)
session_id	string	Yes	Session ID to use

Example Request:

{
  "name": "screenshot",
  "arguments": {
    "full_page": true,
    "format": "png",
    "session_id": "uuid-1234"
  }
}

Example Response:

{
  "type": "image",
  "data": "iVBORw0KGgoAAAANSUhEUgAA...(base64 data)...",
  "mime_type": "image/png"
}

Notes:

• Returns base64-encoded image data
• Automatically resizes if dimensions exceed 8000px (Claude API limit)
• Uses high-quality Lanczos3 interpolation for resizing
• Scrolls element into view before screenshot if selector provided

get_title

Get the title of the current page.

Parameters:

Name	Type	Required	Description
session_id	string	Yes	Session ID to use

Example Request:

{
  "name": "get_title",
  "arguments": {
    "session_id": "uuid-1234"
  }
}

Example Response:

{
  "title": "Example Domain",
  "url": "https://example.com"
}

get_url

Get the current URL of the page.

Parameters:

Name	Type	Required	Description
session_id	string	Yes	Session ID to use

Example Request:

{
  "name": "get_url",
  "arguments": {
    "session_id": "uuid-1234"
  }
}

Example Response:

{
  "url": "https://example.com/page"
}

find_by_text

Find the innermost elements whose visible text contains (or equals) a text. Each result comes with a CSS selector that matches that element only, and a ref usable by every other tool.

Parameters:

Name	Type	Required	Description
text	string	Yes	Text to search for
tag	string	No	HTML tag to search within (default: * for any)
exact	boolean	No	Match the whole visible text, case included (default: false, case-insensitive substring)
multiple	boolean	No	Return all matches or first visible (default: false)
session_id	string	Yes	Session ID to use

Example Request:

{
  "name": "find_by_text",
  "arguments": {
    "text": "Sign In",
    "tag": "button",
    "exact": false,
    "session_id": "uuid-1234"
  }
}

Example Response (Single):

{
  "tag": "button",
  "text": "Sign In",
  "visible": true,
  "selector": "#header > nav > button:nth-of-type(2)",
  "ref": "ref:e42",
  "total_found": 3
}

Example Response (Multiple):

{
  "found": 3,
  "elements": [
    {
      "index": 0,
      "tag": "button",
      "text": "Sign In",
      "visible": true,
      "selector": "#main-login",
      "ref": "ref:e7"
    },
    {
      "index": 1,
      "tag": "a",
      "text": "Sign In Here",
      "visible": true,
      "selector": "#footer > a:nth-of-type(3)",
      "ref": "ref:e8"
    }
  ]
}

Notes:

• Only the innermost match is returned: a wrapper whose descendant also matches (a card, <body>, <html>) is left out
• Text inside <script>, <style> and <template> is never matched
• text is cut at 200 characters; at most 50 elements are returned (found gives the total)
• Prefers visible elements when multiple: false
• ref (e.g. ref:e42) is the same kind of ref as snapshot and works in click, fill_form, get_text, ...