CDP-native MCP server that lets an AI agent drive Chrome at full depth — deep network/console read, robust interaction, multi-session, and API-flow → runnable-code extraction
41 tools with consistent naming (verb_noun pattern) and descriptions present. However, parameter schemas lack depth: most parameters have type and description, but many lack constraints (enums, ranges, patterns). Output schemas are not documented in visible code. Error handling guidance is minimal. Tool descriptions average ~80 chars (below 194-char baseline), missing context on when to use vs similar tools. No tool annotations (readOnlyHint/destructiveHint). Composition is strong (single responsibility per tool, good chaining via IDs), but parameter validation rules and format constraints are underspecified.
Close a browser session or all sessions
Close a specific tab by index
Launch or reuse a Chrome browser session
Recover from a hung page by terminating scripts and disabling page JS
List all active browser sessions with idle time information
List all tabs in the active session
Switch the active session
Switch to a specific tab by index
Clear browser cache
Output schemas not documented. Tools return data but LLMs cannot see expected fields, types, or pagination structure. Forces agents to guess response shape.
Parameter constraints underspecified. 'format' params in page_read, flow_synthesize, flow_export lack enum constraints; 'since' params in net_list, console_errors lack documented valid values. LLMs may pass invalid options.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). LLMs cannot distinguish safe reads from destructive writes without parsing descriptions. Agents may retry destructive ops.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
List cached resources
Get console errors and warnings
Get all console logs
Emulate a device with specific viewport and user agent
Set custom viewport dimensions
Find elements matching criteria
Export captured flow as HAR or other format
Mark the start of an API flow to capture for reverse-engineering
Replay a captured API flow to verify it works
Synthesize captured API flow into runnable code
Get details of a specific network request including full headers
Intercept and modify network requests
List network requests
List pending network requests
Wait for a network request matching criteria
Start observing page changes
Stop observing page changes
Execute multiple page actions in a single batch with optional screenshot
Click an element by badge number or selector
Click at specific coordinates
Execute JavaScript in the page context
Fill an input field by selector or badge
Navigate to a URL
Take a screenshot and analyze the page with badge numbers for clickable elements
Read page content with various options
Take a screenshot of the current page
Get a snapshot of the page DOM and metadata
Type text into a focused input field
Upload a file to a file input
Wait for a condition to be true
Save cookies to a file
Load cookies from a file
Error handling guidance absent. Tools do not document recovery paths (e.g., 'if tab is gone, call browser_tabs to find active tab'). Agents lack actionable next steps on failure.
Descriptions too brief (avg ~80 chars vs 194-char baseline). Missing context on when to use vs similar tools (e.g., page_click vs page_click_at, net_list vs net_pending). LLMs may pick wrong tool.