# MCP Summary Server

An MCP (Model Context Protocol) server for document summarization that keeps full text out of the chat context window.

## Features

- Automatically determines whether to summarize directly or use chunked summarization
- All processing happens server-side
- Returns only the summary to the client
- Configurable chunking parameters
- Bearer token authentication (optional)

## Setup

### Environment Variables

Copy `.env.example` to `.env` and configure:

```bash
cp .env.example .env
```

| Variable | Default | Description |
|----------|---------|-------------|
| PORT | 8080 | HTTP server port |
| API_KEY | (empty) | Bearer token for authentication |
| OPENAPI_URL | http://localhost:8080/v1 | LLM API endpoint |
| OPENAPI_API_KEY | (empty) | LLM API key |
| MODEL_NAME | gpt-4o | LLM model to use |
| LLM_TIMEOUT | 120 | LLM call timeout in seconds |
| CHUNK_SIZE | 4000 | Characters per chunk |
| OVERLAP | 200 | Characters of overlap between chunks |
| TARGET_INTERMEDIATE_SUMMARY_LENGTH | 150 | Words per chunk summary |
| MAX_DIRECT_SUMMARY_LENGTH | 100 | Max final summary length |
| MAX_DIRECT_TEXT_LENGTH | 8000 | Max text length before chunking |

## Running

### Docker

```bash
# Build
docker build -t mcp-summary .

# Run with environment file
docker run -p 8080:8080 --env-file .env mcp-summary

# Run with inline environment variables
docker run -p 8080:8080 \
  -e OPENAPI_URL=http://localhost:8080/v1 \
  -e OPENAPI_API_KEY=your-key \
  -e MODEL_NAME=gpt-4o \
  mcp-summary
```

### Python

```bash
pip install -r requirements.txt
python mcp_summary_server.py
```

## Connecting to OpenWebUI

### In OpenWebUI Admin Settings

1. Go to **Admin Settings → External Tools**
2. Click **+ (Add Server)**
3. Set **Type** to **MCP (Streamable HTTP)**
4. Enter your **Server URL**
5. Set **Authentication**:
   - **None** if no API key is configured
   - **Bearer** if API_KEY is set (provide the key)
6. Save

### Docker Networking

If running both OpenWebUI and MCP Summary in Docker:

```bash
# Use host.docker.internal to reach host machine
docker run -p 8080:8080 \
  -e OPENAPI_URL=http://host.docker.internal:3000/v1 \
  -e OPENAPI_API_KEY=your-key \
  mcp-summary
```

If both containers are on the same Docker network, use the container name directly:

```bash
docker run --network mynetwork -p 8080:8080 \
  -e OPENAPI_URL=http://openwebui-container:8080/v1 \
  -e OPENAPI_API_KEY=your-key \
  mcp-summary
```

## MCP Tool

### summarize_document

Summarizes a document, automatically handling chunking for long text.

**Parameters:**
- `text` (string, required): The document text to summarize
- `max_length` (integer, optional): Maximum summary length in words (default: 100)

**Returns:**
```json
{
  "summary": "The summarized text...",
  "original_length": 12345,
  "method": "direct",  // or "chunked"
  "chunks": 1  // number of chunks used
}
```

## Troubleshooting

### "Failed to connect to MCP server"

1. **Check authentication**: Ensure you haven't selected `Bearer` without a key. Switch to `None` if no token is needed.
2. **Check network connectivity**: Ensure OpenWebUI can reach the MCP server URL
3. **Check LLM connectivity**: Ensure the MCP server can reach the LLM at OPENAPI_URL
4. **Check timeouts**: Increase LLM_TIMEOUT if summarization takes too long

### Infinite loading screen

This may occur if you configured the server as OpenAPI instead of MCP. Fix by:

1. Opening Admin Settings → External Tools
2. Disabling/deleting the problematic connection
3. Re-adding with **Type** set to **MCP (Streamable HTTP)**

### Slow initialization

If the server takes longer than 10 seconds to initialize:
- Increase `MCP_INITIALIZE_TIMEOUT` in OpenWebUI (default: 10 seconds)