​

AI Document Processing: Any Format to MCP

Don’t have an OpenAPI spec? No problem.
Upload any API documentation—PDFs, HTML pages, plain text, even screenshots of documentation—and our AI will analyze it and generate a working MCP server for you.

​

The Magic in 60 Seconds

1. Upload anything → PDF manual, HTML docs, Markdown README, pasted text
2. AI analyzes → Extracts endpoints, parameters, authentication
3. OpenAPI generated → Clean, valid specification created automatically
4. MCP deployed → Working server ready for Claude in under a minute

​

Supported Input Methods

​

📁 File Upload

• Maximum size: 5 MB per file
• Recommended: Under 1 MB for fastest processing

​

📋 Paste Text

• API documentation websites
• Developer portal pages
• README files
• Chat messages or emails describing the API
• Code comments with endpoint details

​

🔗 From URL

• GitHub README files
• GitBook or ReadTheDocs pages
• Swagger UI documentation pages
• Any webpage describing an API

https://github.com/company/repo/blob/main/README.md
https://docs.example.com/api-reference
https://api.example.com/docs/getting-started

• Pages requiring login
• JavaScript-heavy SPAs that don’t render content
• Private documentation behind authentication

​

How It Works

​

Step 1: Upload Your Documentation

1. Click the ”✨ Any Format (AI)” tab
2. Select File Upload, Paste Text, or From URL
3. Provide your documentation

​

Step 2: AI Analysis

• API Title & Description — What the API does
• Base URL — Where the API lives
• Endpoints — All available operations (GET, POST, PUT, DELETE)
• Parameters — Required and optional inputs for each endpoint
• Authentication — API keys, OAuth, Bearer tokens
• Response Schemas — What the API returns

​

Step 3: Review & Confidence Score

• Confidence Score — How certain the AI is about the extraction
  • 🟢 80%+: High confidence, likely accurate
  • 🟡 60-79%: Medium confidence, review suggested
  • 🔴 Below 60%: Low confidence, may need manual adjustment
• Extracted Endpoints — Preview of discovered API operations
• Authentication Type — What auth method was detected

​

Step 4: Generate & Deploy

1. Click “Generate MCP Server”
2. Wait 20-40 seconds for deployment
3. Your MCP server is live and ready for Claude!

​

Your Daily Limits

​

When You’ve Hit Your Limit

1. Wait — Limits reset at midnight in your timezone
2. Use OpenAPI — If you have an OpenAPI spec, upload directly (no limit)
3. Contact us — Need more? Let us know about your use case

​

Best Practices for Success

​

✅ Documentation That Works Great

1. Clear HTTP endpoints
   Copy

   GET /users - Retrieve all users
   POST /users - Create a new user
   GET /users/{id} - Get user by ID
   
2. Explicit parameters
   Copy

   Parameters:
   - page (integer, optional): Page number for pagination
   - limit (integer, optional): Items per page (max 100)
   
3. Base URL specified
   Copy

   Base URL: https://api.example.com/v1
   
4. Authentication documented
   Copy

   Authentication: API Key in header
   Header: X-API-Key: your-api-key
   
5. Response examples
   Copy

   {
     "id": "user_123",
     "email": "user@example.com",
     "created_at": "2024-01-15T10:30:00Z"
   }
   

​

⚠️ Documentation That May Struggle

• SDK documentation only — Code samples without REST endpoints
• Very vague descriptions — “Call the API to get data”
• No endpoint URLs — Just function names without HTTP methods
• Marketing materials — Product pages without technical details
• Heavily image-based docs — Screenshots instead of text

​

🎯 Quick Improvements

## API Reference

Base URL: https://api.yourservice.com/v1

Authentication: Bearer token in Authorization header

### Endpoints

#### GET /items
Get all items.
- Query params: limit (int), offset (int), status (string)
- Returns: Array of item objects

#### POST /items
Create a new item.
- Body: { "name": string, "price": number }
- Returns: Created item object

#### GET /items/{id}
Get item by ID.
- Path param: id (required)
- Returns: Item object

​

Troubleshooting

​

”Documentation quality insufficient”

1. Add more detail — Include explicit endpoints with HTTP methods
2. Specify the base URL — Tell the AI where the API lives
3. List parameters — Name each parameter with its type
4. Try different input — Upload a different doc or paste text directly

​

”Daily limit reached”

1. Wait for reset — Limits reset at midnight (your timezone)
2. Use OpenAPI instead — Direct OpenAPI upload has no daily limit
3. Convert manually — Use Swagger Editor to create a spec

​

”AI service temporarily at capacity”

1. Wait a moment — Try again in 30-60 seconds
2. Try during off-peak hours — Early morning or late evening
3. Use OpenAPI directly — Bypass AI by uploading an existing spec

​

”File too large”

1. Extract key sections — Copy/paste the API reference section only
2. Use URL instead — Link to documentation rather than uploading
3. Split the document — Process large APIs in sections

​

Low Confidence Score (Below 60%)

1. Review the preview — Check if endpoints look correct
2. Add missing info — Edit the generated spec before deploying
3. Provide cleaner input — Remove marketing content, focus on technical details
4. Try a different format — Some formats work better than others

​

Frequently Asked Questions

​

Can I edit the generated OpenAPI spec?

• Review extracted endpoints
• Verify parameter types
• Check authentication settings

​

What if the AI misses some endpoints?

1. Deploy what was found first
2. Manually create a complete OpenAPI spec for missing endpoints
3. Create a second deployment with the additional endpoints

​

Does AI processing count against my rate limits?

​

Can I process multiple APIs at once?

1. Process each API separately
2. Create individual MCP deployments
3. Configure each in Claude Desktop

​

What data does the AI see?

• Your uploaded documentation content
• Any URLs you provide for fetching

​

Is there a way to improve AI accuracy?

1. Structured documentation — Clear headings, tables, code blocks
2. Technical focus — API reference sections rather than marketing
3. Explicit examples — Request/response samples help a lot
4. Standard terminology — REST conventions, HTTP method names

​

Examples: What Works Best

​

Great Input: GitHub README

# Payments API

Base URL: `https://api.payments.io/v2`

## Authentication
Include your API key in the header:

## Endpoints

### Create Payment
`POST /payments`

**Request Body:**
```json
{
  "amount": 1000,
  "currency": "USD",
  "customer_id": "cus_123"
}

​

Get Payment

**Result:** High confidence, accurate extraction ✅

---

### Okay Input: Text Description

**Result:** Medium confidence, basic extraction ⚠️

---

### Difficult Input: Vague Description

**Result:** Low confidence, insufficient detail ❌

---

## Still Need Help?

- 📚 **[Troubleshooting Guide](/support/troubleshooting)** — General platform issues
- ❓ **[FAQ](/support/faq)** — Common questions answered
- 📖 **[User Guide](/guides/user-guide)** — Complete platform documentation
- 💬 **[Contact Support](https://tydli.io/contact)** — We're here to help!

---

*AI Document Processing is continuously improving. Have feedback? We'd love to hear what documentation formats work well for you.*