Local loopback bridge between the DevScope Chrome extension and the Claude / Cursor CLI, providing MCP servers for browser control, task management, WhatsApp, Gmail, Calendar, and Meta Ads integration
DevScope Bridge provides 22 browser automation tools with consistent naming (browser_* prefix) and descriptions. All tools have input schemas with type definitions and parameter descriptions. However, descriptions are brief (avg ~80 chars, well below the 194-char baseline for A+ tools), lack actionable context about when to use each tool, and omit error recovery guidance. Output schemas are not documented. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications. Parameter descriptions are minimal, e.g., 'Optional tab ID; uses bound tab if not provided' repeats across 20 tools without explaining what 'bound tab' means or how to bind one. Missing dependency hints (e.g., 'Call browser_list_tabs first to find a tab_id'). Error handling is present but generic.
Assert that text is present on the page, optionally within a specific element
Click an element matching the selector or referenced by element ref
Execute JavaScript code in the page context and return the result
Fill an input or textarea element with text
Focus a specific browser tab
Get console logs from the page
Query elements matching a selector and return their properties
Get network request history and performance metrics for the page
Descriptions are too brief (avg ~80 chars vs 194-char baseline). Most lack context about when to use the tool, prerequisites, or what it returns. E.g., 'Get current page information' does not explain what 'current' means or when to call this vs browser_snapshot.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite explicit Risk classifications. browser_click, browser_fill, browser_navigate are marked WRITE but lack destructiveHint annotation. Agents cannot infer safety without annotations.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
Get current page information including URL, title, viewport dimensions, and scroll position
Highlight elements on the page with optional labels
Hover over an element
Press a keyboard key, optionally on a specific element
List all Chrome windows and tabs, showing which tab is bound to the current session
Navigate to a URL in the specified or bound tab
Open a new browser tab or window with the specified URL
Capture a screenshot of the current or specified tab
Scroll an element into view
Select an option in a select element
Generate an accessibility tree snapshot of the page with interactive elements
Upload a file to a file input element
Wait for the page to navigate to a new URL
Wait for an element matching the selector to appear on the page
Parameter descriptions are repetitive and lack actionable detail. 'Optional tab ID; uses bound tab if not provided' appears in 20 tools but never explains what 'bound tab' means, how to bind one, or what happens if no tab is bound. Missing dependency hints like 'Call browser_list_tabs first to find a tab_id'.
Output schemas are not documented. Tools like browser_snapshot, browser_get_elements, browser_get_network, and browser_get_console return complex structures (accessibility trees, network metrics, console logs) but no schema is visible. LLMs cannot plan downstream calls without knowing response structure.
browser_evaluate accepts arbitrary JavaScript with no validation or safety guardrails. Description does not warn about injection risks or side effects. No mention of what happens if the expression throws an error or times out.