> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tydli.io/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Document Processing

> Convert any API documentation to MCP servers with AI

# 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

No coding. No manual conversion. Just upload and go.

***

## Supported Input Methods

### 📁 File Upload

| Format         | Extension       | Best For                                   |
| -------------- | --------------- | ------------------------------------------ |
| **PDF**        | `.pdf`          | Official API documentation, generated docs |
| **Word**       | `.docx`         | Internal API documentation                 |
| **Markdown**   | `.md`           | GitHub READMEs, developer docs             |
| **Plain Text** | `.txt`          | Simple API references, notes               |
| **JSON**       | `.json`         | Existing partial specs, config files       |
| **YAML**       | `.yaml`, `.yml` | Kubernetes configs, partial OpenAPI        |
| **HTML**       | `.html`         | Saved web documentation                    |

**File limits:**

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

***

### 📋 Paste Text

Copy and paste documentation directly from:

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

**Pro tip:** When pasting from websites, include the base URL and any authentication details you see.

***

### 🔗 From URL

Enter a URL to any publicly accessible API documentation:

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

**Works with:**

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

**Doesn't work with:**

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

***

## How It Works

### Step 1: Upload Your Documentation

Choose your input method and provide your API documentation:

!\[Input Selection]

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

### Step 2: AI Analysis

Our AI reads your documentation and extracts:

* **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

The AI shows you what it found:

* **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

If everything looks good:

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

| What           | Limit                    | Resets                     |
| -------------- | ------------------------ | -------------------------- |
| AI Conversions | **5 per day**            | Midnight (your local time) |
| File Size      | **5 MB** max             | —                          |
| Content Length | **\~400,000 characters** | —                          |

### When You've Hit Your Limit

The button will show "Limit Reached" with a countdown to reset.

**Options:**

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

The AI excels when your documentation includes:

1. **Clear HTTP endpoints**
   ```
   GET /users - Retrieve all users
   POST /users - Create a new user
   GET /users/{id} - Get user by ID
   ```

2. **Explicit parameters**
   ```
   Parameters:
   - page (integer, optional): Page number for pagination
   - limit (integer, optional): Items per page (max 100)
   ```

3. **Base URL specified**
   ```
   Base URL: https://api.example.com/v1
   ```

4. **Authentication documented**
   ```
   Authentication: API Key in header
   Header: X-API-Key: your-api-key
   ```

5. **Response examples**
   ```json theme={null}
   {
     "id": "user_123",
     "email": "user@example.com",
     "created_at": "2024-01-15T10:30:00Z"
   }
   ```

### ⚠️ Documentation That May Struggle

The AI might have difficulty with:

* **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

If the AI confidence is low, try adding:

```markdown theme={null}
## 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"

**What it means:** The AI couldn't extract enough structured API information.

**How to fix:**

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"

**What it means:** You've used your 5 free AI conversions today.

**How to fix:**

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](https://editor.swagger.io) to create a spec

### "AI service temporarily at capacity"

**What it means:** Our AI processing system is experiencing high demand.

**How to fix:**

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"

**What it means:** Your file exceeds the 5 MB limit.

**How to fix:**

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%)

**What it means:** The AI is uncertain about what it extracted.

**How to fix:**

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?

Yes! After AI analysis, you'll see a preview of the generated specification. You can:

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

For advanced edits, copy the generated JSON and modify it before deploying.

### What if the AI misses some endpoints?

If endpoints are missing:

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

Or, enhance your input documentation with clearer endpoint definitions and try again.

### Does AI processing count against my rate limits?

The AI conversion uses your **daily AI quota** (5 per day), which is separate from your MCP server request limits. Once deployed, MCP servers use your normal request quota.

### Can I process multiple APIs at once?

Currently, AI processing works on one document at a time. For multiple APIs:

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

### What data does the AI see?

The AI processes:

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

We do **not** store your documentation after processing. Generated OpenAPI specs are saved only in your deployment.

### Is there a way to improve AI accuracy?

Yes! The AI works best with:

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

```markdown theme={null}
# Payments API

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

## Authentication
Include your API key in the header:
```

X-API-Key: your\_key\_here

````

## Endpoints

### Create Payment
`POST /payments`

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

### Get Payment

`GET /payments/{id}`

Returns payment details.

```

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

---

### Okay Input: Text Description

```

Our API lets you manage users. You can create users with POST to /api/users,
get a user with GET /api/users/id, and delete with DELETE /api/users/id.
The base URL is [https://myapp.com](https://myapp.com). Use an API key in the X-Key header.

```

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

---

### Difficult Input: Vague Description

```

The system provides user management capabilities through our RESTful interface.
Contact support for API access credentials.

```

**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.*
```