MCP server for stealth browser automation via Camoufox — 133 tools, plus a browserless TLS-impersonating HTTP path
40 tools with consistent naming (verb_noun pattern) and present descriptions, but critical gaps in schema completeness and parameter documentation. Most tools lack explicit input schema definitions in visible source code, schemas are inferred from parameter lists rather than formally declared. Descriptions average ~80 chars (below 194-char baseline), many under 100 chars. No output schemas documented. Error handling absent. Security concerns: browser_launch accepts headless/fresh_profile booleans with no validation guidance; evaluate executes arbitrary JS with no sandboxing warning. Composition is reasonable (single responsibility per tool), but parameter descriptions lack format/constraint details (e.g., 'URL to navigate to' vs 'URL (http/https, max 2048 chars)').
Capture authentication state
Close the active browser instance
Launch a Camoufox browser instance with optional URL navigation
Capture a snapshot of the current page with element references
Check a checkbox element
Click an element by reference
Click an element by visible text
Delete a browser cookie
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., what does browser_snapshot return? What fields are in each element reference?).
evaluate tool executes arbitrary JavaScript with no sandboxing warning or security guidance. Description lacks mention of code injection risks or validation requirements.
Parameter descriptions lack format/constraint details. E.g., 'URL to navigate to' does not specify http/https only, max length, or validation rules. 'Text to search for' does not specify max length or regex support.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 55 | 2026-07-28+ | v2 |
Export cookies to a file
Get browser cookies
Set a browser cookie
Detect repeating content patterns on the page
Automatically handle all dialogs
Handle a browser dialog (alert, confirm, prompt)
Execute JavaScript in the page context
Extract structured data from repeating elements
Extract a table as JSON
Fill an input field with text
Find elements by visible text
Audit browser fingerprint for inconsistencies
Get all links from the page
Make an HTTP request (browserless)
Get cookies for HTTP requests
Inspect an element's properties
Start intercepting network requests
Stop intercepting network requests
Navigate to a URL in the current page
Reset the browser profile to a clean state
Scrape page as markdown (browserless)
Scrape page content as JSON
Take a screenshot of the current page
Search using a search engine endpoint
Select an option in a select element
Fetch a URL, escalating to browser if blocked
Save browser storage state to a file
Close a tab
List all open tabs
Open a new tab
Switch to a specific tab
Uncheck a checkbox element
No error handling guidance. Tools lack recovery hints (e.g., 'Element not found. Try browser_snapshot() to see available elements'). No categorization of retryable vs fatal errors.
Destructive operations (reset_profile, cookie_delete, storage_state_save) lack confirmation or dry-run support. No warning in descriptions that these are irreversible.