hosaypeng Full Page Analysis
Retrieve and analyze webpage content with full fidelity, bypassing WebFetch's summarization.
Mode Selection
Choose the most token-efficient approach based on need:
| Flag | Method | Best for |
|---|---|---|
| (default) | Smart routing | Auto-selects based on content |
--targeted |
JavaScript extraction | Known section/selector needed |
--full |
Complete text dump | Legal docs, specs, full context |
Process
Default Mode (Smart Routing)
- Quick probe first: Use WebFetch with prompt "List all section headings and estimate word count"
- Decide extraction method:
- Short page (<2000 words): Use get_page_text directly, no save
- Long page + specific focus: Use JavaScript to extract relevant section
- Long page + general analysis: Full extraction with chunking
Targeted Mode (--targeted)
Skip full extraction. Use JavaScript to pull specific content:
// Example: Extract just the pricing section
document.querySelector('[data-section="pricing"]')?.innerText ||
document.querySelector('#pricing')?.innerText ||
[...document.querySelectorAll('h2')].find(h => h.textContent.includes('Pricing'))?.parentElement?.innerText
Use mcp__claude-in-chrome__javascript_tool with a selector based on the focus area.
Full Mode (--full)
- Get tab context (reuse existing tab if same domain)
- Navigate + get_page_text in sequence
- Only save to scratchpad if content >8000 words (for chunked re-reading)
- Analyze in single pass when possible
Token Optimization Rules
- Reuse tabs: Check tabs_context first, don't create new tabs for same domain
- Skip scratchpad: Only write to file if content requires multiple passes
- Parallel calls: Combine independent operations
- Early exit: If WebFetch probe answers the question, stop there
Output
Provide a structured analysis:
## Source
[URL] | Fetched: [timestamp]
## Summary
[2-3 sentence overview]
## Key Details
- [Specific finding with exact quotes where relevant]
- [Another finding]
## Structure
[How the page is organized - sections, navigation, etc.]
## [Focus Area] (if specified)
[Detailed analysis of the user's area of interest]
## Extraction Method
[Which mode was used and why]
Arguments
- url (required): The webpage to analyze
- focus area (optional): Specific aspect to emphasize (e.g., "pricing", "technical specs", "author bio")
Constraints
- If the page requires authentication, inform the user
- Respect robots.txt and rate limits
- Don't reproduce copyrighted content verbatim in the response - use short quotes
- The saved file in scratchpad is for YOUR analysis, not for giving to the user
When to Use This vs WebFetch
Use /analyze-page when you need:
- Exact quotes or specific wording
- Complete table/list data
- Full context before making decisions
- To analyze structure or navigation
- Legal, technical, or specification content
Use WebFetch when you need:
- Quick factual lookups
- General understanding of a topic
- Multiple pages scanned rapidly
Troubleshooting
- Page returns empty content or 403: The page may require authentication or block automated access. Fall back to
mcp__claude-in-chrome__navigate+get_page_text. If auth is required, inform the user. - Chrome MCP tools unavailable: Fall back to WebFetch-only mode. Note in the output that full-fidelity extraction was not possible.
- Page content exceeds context window: Use targeted mode to extract specific sections, or process in chunks.