Claude analysing a page and providing actionable GEO recommendations

Talk naturally to Claude and get instant GEO analysis - no switching contexts, no external tools, no manual work.

Three Powerful Tools at Your Command:

• analyze_url: Deep-dive analysis of any webpage with extractability scoring, semantic analysis, and prioritised recommendations
• compare_extractability: Side-by-side comparison of up to 5 URLs to benchmark against competitors
• validate_rewrite: Before/after comparison proving your content improvements with percentage changes

🔍 Content Extraction: Jina.ai Reader API strips away clutter and returns pristine markdown - 1M tokens free per month
🤖 Semantic Analysis: Cloudflare Workers AI with Llama 3.3 70B extracts semantic triples, entities, and relationships
⚙️ Pattern Analysis: Custom regex-based scoring measures claim density, sentence structure, and temporal markers
🔌 MCP Protocol: Native integration with Claude Desktop via Anthropic's Model Context Protocol

Extractability Scoring (0-10)

• Sentence length optimisation (target: 15-20 words)
• Claim density measurement (target: 4+ per 100 words)
• Temporal marker detection for freshness signals

Semantic Intelligence

• Subject-predicate-object triple extraction
• Named entity recognition (PERSON, ORG, PRODUCT, etc.)
• Entity diversity and relationship mapping

Actionable Recommendations

• Prioritised by impact (high/medium/low)
• Specific location references with line numbers
• Before/after examples with clear rationale

1. Click "Deploy to Cloudflare" above ☝️
2. Sign in to your Cloudflare account
3. Fork repository (done automatically)
4. Worker deploys automatically to your account
5. Copy your Worker URL (shown after deployment)
6. Configure Claude Desktop - Add to your config:

   {
     "mcpServers": {
       "geo-analyzer": {
         "command": "npx",
         "args": ["-y", "@houtini/geo-analyzer"],
         "env": {
           "GEO_WORKER_URL": "https://geo-analyzer.YOUR-SUBDOMAIN.workers.dev"
         }
       }
     }
   }

7. Restart Claude Desktop
8. Test it: Ask Claude "Analyse https://example.com for 'your topic'"

See detailed manual setup guide below ↓

Before deploying, ensure you have:

• ✅ Cloudflare Account - Free tier is sufficient (Sign up)
• ✅ Claude Desktop - With MCP support (Download)
• ✅ Node.js 20+ - For local development (Download)
• 🎯 Jina AI API Key - Highly recommended for production use (Get free key)

Why Jina.ai? The Reader API transforms messy web pages into clean, AI-ready markdown—removing ads, navigation, footers, and clutter. This dramatically improves analysis accuracy.

Free Tier Benefits:

• 1M tokens per month (roughly 250 page analyses per day)
• Clean markdown output optimised for LLM processing
• Automatic image caption extraction
• Link summary generation
• Works with any publicly accessible URL

Without Jina: Analysis still functions but uses basic content extraction. For production use and best results, the Jina API key is strongly recommended.

Generative Engine Optimisation (GEO) is the practice of optimizing content to appear in AI-generated responses from systems like ChatGPT, Claude, Perplexity and other LLM-based search engines.

This tool implements methodologies from "GEO: Generative Engine Optimization" by Aggarwal et al., published at ACM SIGKDD 2024 (read the full paper). The research analysed 10,000 diverse queries across multiple AI systems to identify which content optimisation techniques most effectively improve visibility in AI-generated responses.

Core Principles from the Research:

Extractability: How easily AI systems can extract discrete facts and claims from content. The study found that content with higher claim density (4+ claims per 100 words) and optimal sentence length (15-20 words) performed significantly better.

Citability: How quotable and attributable the content is through semantic triples and entity relationships. Pages with rich entity diversity and clear subject-predicate-object relationships were 40% more likely to be cited by AI systems.

Readability: How well-structured the content is for AI parsing through clear headings, lists, and logical chunking. Hierarchical structure improved AI comprehension and extraction accuracy.

Key Finding: The research evaluated 9 optimisation methods and found that traditional SEO approaches like keyword stuffing performed poorly (often reducing visibility). Instead, methods focused on extractability (claim density, semantic triples, entity extraction) delivered the strongest improvements, increasing AI citation rates by up to 40% compared to baseline content.

This MCP server provides methodology-based analysis, not performance predictions. We measure factors the research identified as important for AI citability, helping you optimise content according to proven principles.

Important: Results show what the analysis found, not guarantees about AI system behaviour. Use this as a guide for content optimisation, not a prediction tool.

This tool requires you to deploy your own Cloudflare Worker. There is no public API endpoint.

• No Storage: Content is analysed in real-time, nothing is persisted
• No Tracking: No analytics or usage tracking
• Your Infrastructure: Runs on your Cloudflare account
• Your Logs: Only you can access Worker logs and metrics

graph LR
    A[Claude Desktop] -->|MCP Protocol| B[MCP Server]
    B -->|HTTPS| C[Your Cloudflare Worker]
    C -->|Content Fetch| D[Jina AI API]
    C -->|Semantic Analysis| E[Workers AI - Llama 3.3]
    E -->|Scores & Recommendations| F[GEO Analysis Results]
    F -->|JSON Response| A

• MCP Server (packages/mcp-server): Claude Desktop integration via Model Context Protocol
• Cloudflare Worker (packages/cloudflare-worker): Serverless API with Workers AI
• Shared Types (packages/shared): TypeScript definitions

• Node.js 20+ and npm
• Claude Desktop with MCP support
• Cloudflare account (free tier works)
• Jina AI API key (optional, free tier: https://jina.ai/)

Option A: Install from NPM (when published):

npm install -g @houtini/geo-analyzer

Option B: Clone for Development:

git clone https://github.com/houtini/geo-analyzer.git
cd geo-analyzer
npm install
npm run build

cd packages/cloudflare-worker

# Login to Cloudflare (first time only)
npx wrangler login

# Deploy your worker
npx wrangler deploy

Save the Worker URL displayed after deployment. It will look like:

https://geo-analyzer.YOUR-SUBDOMAIN.workers.dev

Location of config file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Add this configuration:

{
  "mcpServers": {
    "geo-analyzer": {
      "command": "npx",
      "args": ["-y", "@houtini/geo-analyzer"],
      "env": {
        "GEO_WORKER_URL": "https://geo-analyzer.YOUR-SUBDOMAIN.workers.dev",
        "JINA_API_KEY": "your-jina-key-here"
      }
    }
  }
}

Configuration Variables:

Without Jina API Key:

• Analysis still works but uses basic content extraction
• Jina Reader provides cleaner, more accurate content parsing
• Recommended for production use

Close and reopen Claude Desktop completely for the MCP server to connect.

Ask Claude:

Analyze https://example.com for the query "best practices"

Claude will use the GEO Analyser tool to provide detailed extractability metrics and recommendations.

Analyze https://yoursite.com/blog/article for "content marketing"

Compare GEO scores for these URLs:
- https://site1.com/page
- https://site2.com/page
- https://site3.com/page

Target query: "your topic"

I rewrote this content. Compare it to the original at https://site.com/original

[Your optimized content here]

Target query: "your topic"

The tool supports multiple Cloudflare AI models:

Usage: Pass aiModel parameter in API requests or tool calls.

• Overall Score: Combined weighted average
• Extractability: How easily AI can pull facts
• Readability: How well-structured for AI parsing
• Citability: How quotable and attributable

• Sentence length distribution
• Claim density measurements
• Semantic triple identification
• Entity extraction and diversity
• Temporal marker detection
• Content chunking analysis
• Query alignment scoring

• Prioritized by impact (high/medium/low)
• Specific location references
• Before/after examples
• Clear rationale for each suggestion

Edit packages/cloudflare-worker/wrangler.toml:

name = "my-geo-analyzer"
main = "src/index.ts"
compatibility_date = "2025-01-01"

[observability]
enabled = true

[ai]
binding = "AI"

cd packages/mcp-server
npm run dev

# Test MCP server locally
cd packages/mcp-server
npm run build
node dist/index.js

# Test Worker locally
cd packages/cloudflare-worker
npx wrangler dev

Symptoms: Error during wrangler deploy

Solutions:

1. Ensure you're logged in: npx wrangler login
2. Check Cloudflare account is active and verified
3. Verify Node.js version: node --version (should be 20+)
4. Check for typos in wrangler.toml
5. Try deploying again - Cloudflare may be provisioning resources

Symptoms: Tools don't appear in Claude Desktop

Solutions:

1. Restart Claude Desktop completely (close all windows)
2. Check config file location:
   • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
   • Windows: %APPDATA%\Claude\claude_desktop_config.json
3. Verify JSON syntax is valid (use JSONLint)
4. Check GEO_WORKER_URL has no trailing slash
5. Ensure Worker URL is accessible (open in browser - should show {"error":"Not found"})
6. Look for errors in Claude Desktop logs

Symptoms: Analysis fails with authentication error

Solutions:

1. Check JINA_API_KEY is correctly set in Claude config
2. Verify API key at Jina AI dashboard
3. Ensure no extra spaces or quotes in API key
4. Try without Jina API key (analysis will use basic extraction)

Symptoms: Long wait times or timeout errors

Solutions:

1. Check page is publicly accessible (not behind login/paywall)
2. Try a smaller/simpler page first
3. Verify Cloudflare Workers AI is enabled in your account
4. Check Cloudflare dashboard for any service issues
5. Consider the page size - very large pages may take longer

Symptoms: Error about 1MB limit

Solutions:

1. This only affects the validate_rewrite tool
2. Content must be under 1MB (roughly 250,000 words)
3. For normal URL analysis, there's no size limit
4. Try analyzing the original URL instead

Symptoms: MCP server fails to start

Solutions:

1. URL must include https:// protocol
2. URL must be a .workers.dev domain or custom domain
3. No trailing slashes
4. Copy exactly from Wrangler deployment output

If you're stuck:

1. Check Worker logs: npx wrangler tail in worker directory
2. Test Worker directly: Visit Worker URL in browser
3. Verify MCP connection: Look for "Connected" status in Claude
4. Review security audit: See SECURITY-AUDIT.md for security best practices

• No public endpoints: You deploy your own private Cloudflare Worker
• API key protection: Jina API key handled securely via environment variables
• Input validation: All inputs validated and sanitised
• Content size limits: Maximum 1MB per analyze request
• Rate limiting: Cloudflare provides automatic DDoS protection
• HTTPS only: All communication encrypted via TLS

# Never commit your Worker URL to version control
# Add to .gitignore:
echo "claude_desktop_config.json" >> .gitignore

# Regularly audit dependencies
npm audit
npm audit fix

# Update dependencies
npm update

# Check your Cloudflare Workers usage
npx wrangler metrics

For detailed security information, see SECURITY-AUDIT.md.

• Multi-model support: GPT-4, Claude, Gemini analysis
• Batch processing: Analyze entire sitemaps
• Historical tracking: Track GEO scores over time
• Competitor monitoring: Automated competitor analysis
• WordPress plugin: Direct CMS integration
• Chrome extension: Browser-based analysis
• A/B testing: Compare optimization approaches

• Python SDK for programmatic access
• CI/CD integration for content pipelines
• Slack/Discord notifications
• Custom scoring weights
• Export reports to PDF/CSV

Want to contribute? See CONTRIBUTING.md or open an issue!

See full pricing details at Cloudflare and Jina AI.

This project is built on research from:

Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., & Deshpande, A. (2024).
GEO: Generative Engine Optimization.
ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD '24)
DOI: 10.1145/3637528.3671900 | arXiv: 2311.09735

We thank the researchers for making their findings publicly available and contributing to the understanding of content optimisation for AI systems.

Built with ❤️ for the creator economy by Houtini

Contact: [email protected]