Skip to main content

​
MCP Tools Reference

The BAP MCP server exposes 23 tools organized into six categories. With --slim mode, only 5 essential tools are exposed.
Navigate back in browser history.No parameters.
Navigate forward in browser history.No parameters.
Reload the current page.No parameters.

​
Element Interaction

Click an element using semantic selectors.
​
selector
string
required
Element selector (see Selector Formats below).
Fill a form field. Clears existing content first.
​
selector
string
required
Element selector targeting the input field.
​
value
string
required
The value to fill.
Type text character by character. First clicks the element, then types. Triggers individual key events.
​
selector
string
required
Element selector.
​
text
string
required
Text to type character by character.
​
delay
number
Delay in milliseconds between keystrokes.
Press keyboard keys (Enter, Tab, shortcuts).
​
key
string
required
Key to press (e.g., Enter, Tab, Escape, Control+a).
​
selector
string
Optional element to focus before pressing.
Select an option from a dropdown.
​
selector
string
required
Element selector targeting the <select> element.
​
value
string
required
Option value or visible text to select.
Scroll the page or a specific element.
​
direction
string
default:"down"
Scroll direction: up, down, left, right.
​
amount
number
default:"300"
Pixels to scroll.
​
selector
string
Optional element to scroll into view.
Hover over an element.
​
selector
string
required
Element selector.

​
Observation

AI-optimized page observation. Returns interactive elements with stable refs. Supports incremental diffs and response tiers.
​
maxElements
number
default:"50"
Maximum elements to return.
​
filterRoles
string[]
Filter by ARIA roles (e.g., ["button", "link", "textbox"]).
​
includeScreenshot
boolean
default:"false"
Include an annotated screenshot with element refs.
​
incremental
boolean
default:"false"
Return only changes since the last observation: { added, updated, removed }.
​
responseTier
string
Response tier: full, interactive, minimal.
​
includeWebMCPTools
boolean
default:"false"
Include tools discovered via the WebMCP standard.
observe({ maxElements: 20, responseTier: "interactive", includeScreenshot: true })
Take a screenshot of the current page.
​
fullPage
boolean
default:"false"
Capture the entire scrollable page.
Token-efficient YAML accessibility snapshot. Approximately 80% fewer tokens than accessibility.No parameters.
Prefer aria_snapshot over accessibility for cost efficiency. Same structure, far fewer tokens.
Full accessibility tree.No parameters.
Get page text content.
​
format
string
default:"text"
Output format: text or markdown.
Query element properties: exists, visible, enabled, checked, value.
​
selector
string
required
Element selector.

​
AI Agent Methods

Execute a sequence of browser actions in a single call. Supports fused pre/post observation.
​
steps
array
required
Array of action steps. Each step has action, selector, and optionally value.
steps: [
  { action: "action/fill", selector: "label:Email", value: "user@example.com" },
  { action: "action/click", selector: "role:button:Submit" }
]
​
preObserve
object
Observe page state before executing steps.
​
postObserve
object
Observe page state after executing steps. Contains maxElements, responseTier.
act({
  steps: [
    { action: "action/fill", selector: "@e1", value: "user@example.com" },
    { action: "action/fill", selector: "@e2", value: "password123" },
    { action: "action/click", selector: "role:button:Sign in" }
  ],
  postObserve: { responseTier: "interactive" }
})
Extract structured data from the page using schema and CSS heuristics.
​
instruction
string
Natural language description of what to extract.
​
mode
string
Extraction mode: single (one object) or list (array of objects).
​
schema
object
JSON Schema defining the extraction shape.
extract({
  instruction: "Extract product listings",
  mode: "list",
  schema: {
    type: "array",
    items: {
      type: "object",
      properties: {
        name: { type: "string" },
        price: { type: "number" }
      }
    }
  }
})

​
Page Management

List all open pages/tabs.No parameters.
Switch to a different page/tab.
​
pageId
string
required
Page ID from the pages result.
Close the current page/tab.No parameters.

​
Discovery

Discover tools that websites expose via the W3C WebMCP standard. Checks for declarative (form-based) and imperative (navigator.modelContext) tool definitions.No parameters.

​
Selector Formats

All tools that accept a selector parameter support these formats: 
role:button:Submit        # ARIA role + accessible name (recommended)
text:Sign in              # Visible text content
label:Email address       # Associated label
testid:submit-button      # data-testid attribute
ref:@submitBtn            # Stable element reference from observe
css:.btn-primary          # CSS selector (fallback)
xpath://button[@type]     # XPath selector (fallback)
Always prefer role: selectors for buttons, links, and inputs. They survive DOM changes. Never copy CSS selectors from browser DevTools.

​
Resources

The MCP server also exposes 4 resources:
ResourceDescription
bap://page/stateCurrent page URL and title
bap://page/accessibilityFull accessibility tree
bap://page/aria-snapshotYAML accessibility snapshot
bap://page/screenshotCurrent page screenshot