README.md•7.63 kB
# Meeting Transcript Analyzer - Multi-Agent MCP App
A multi-agent system that analyzes meeting transcripts using AI-powered summarization, key point extraction, and task identification. **Fully compliant with the Model Context Protocol (MCP) standard.**
## Features
- **Modern Web UI**: Clean, responsive horizontal layout interface for easy interaction (see [screenshot](#welcome-page))
- **Summarize**: Generate concise summaries of meeting transcripts (see [screenshot](#summarize-flow))
- **Key Highlights**: Extract and display key points as bullet points (•) (see [screenshot](#key-highlights-flow))
- **Grab Tasks**: Identify actionable tasks from meeting discussions (see [screenshot](#task-extraction-flow))
- **MCP Compliant**: Full adherence to Model Context Protocol standards
## MCP Compliance
This application is fully compliant with the **Model Context Protocol (MCP)** standard as defined in the [official MCP specification](https://modelcontextprotocol.io/specification):
### ✅ MCP Standard Implementation
- **Discovery Endpoint**: `/.well-known/mcp.json` (MCP compliant)
- **Tool Schema**: JSON Schema format with `inputSchema` property
- **Response Format**: MCP-compliant `content` array with structured content
- **Request Format**: Standard MCP `invoke` endpoint with `name` and `arguments`
- **Capabilities**: Proper capabilities object in discovery response
### 🔧 MCP Endpoints
Each agent exposes MCP-compliant endpoints:
```
GET /.well-known/mcp.json # Tool discovery (MCP standard)
POST /invoke # Tool invocation (MCP standard)
GET /health # Health check
```
### 📋 MCP Tool Definitions
**Summarizer Agent Tools:**
- `summarize_transcript`: Generates concise meeting summaries
- `highlight_key_points`: Extracts 3-5 key insights as bullet points
**Task Extractor Agent Tools:**
- `extract_tasks`: Identifies actionable tasks from transcripts
### 🔄 MCP Response Format
All tools return MCP-compliant responses:
```json
{
"content": [
{
"type": "text",
"text": "Tool output content here..."
}
]
}
```
## Prerequisites
- Python 3.9+
- OpenAI API key
## Setup
1. **Install dependencies:**
```bash
pip install -r requirements.txt
```
2. **Configure OpenAI API Key:**
Create a file named `openai_key.txt` in the project root and add your OpenAI API key:
```
sk-your-openai-api-key-here
```
## Running the Agents
**Important:** All commands must be run from the project root directory.
**Note:** Start the sub-agents first, then the super agent to ensure proper tool registration.
### 1. Start the Summarizer Agent (Port 8001)
```bash
python3 -m uvicorn agents.summarizer_agent:summarizer_app --reload --port 8001
```
### 2. Start the Task Extractor Agent (Port 8002)
```bash
python3 -m uvicorn agents.task_extractor_agent:task_app --reload --port 8002
```
### 3. Start the Super Agent (Port 8000)
```bash
python3 -m uvicorn agents.super_agent:super_app --reload --port 8000
```
## Using the Application
1. **Access the Web Interface:**
Open your browser and go to: [http://localhost:8000](http://localhost:8000)
2. **Analyze a Transcript:**
- Paste your meeting transcript in the left textarea
- Enter a prompt like "Summarize this meeting" or "Extract key points" in the second textarea
- Click "Analyze Transcript"
3. **View Results:**
- Results appear in the right panel with structured formatting
- Summaries appear as formatted paragraphs
- Key points display as clean bullet points (•)
- Tasks show as numbered actionable items
- Metadata shows transcript length, tool used, and point/task counts
## Application Screenshots
#### Welcome Page
_The clean, modern interface users see when first opening the application._
#### Summarize Flow
_The application summarizing a meeting transcript with a brief, concise style._
#### Key Highlights Flow
_Extracting key insights and main points from a meeting transcript as bullet points._
#### Task Extraction Flow
_Identifying and extracting actionable tasks from meeting discussions._
## Architecture
- **Super Agent (Port 8000)**: Main entry point that serves the web UI and orchestrates sub-agents
- **Summarizer Agent (Port 8001)**: Handles transcript summarization and key point extraction
- **Task Extractor Agent (Port 8002)**: Identifies and extracts actionable tasks from transcripts
## Technical Details
- **Backend Formatting**: All response formatting is handled by the super agent for consistent UI presentation
- **Structured Responses**: Responses include type, title, content, and metadata fields
- **MCP Protocol**: Uses Model Context Protocol for agent communication
- **Responsive Design**: UI adapts to mobile devices with vertical stacking
## MCP Integration
### For MCP Clients
To integrate with MCP-compliant clients:
1. **Discover Tools:**
```bash
curl http://localhost:8001/.well-known/mcp.json
curl http://localhost:8002/.well-known/mcp.json
```
2. **Invoke Tools:**
```bash
curl -X POST http://localhost:8001/invoke \
-H "Content-Type: application/json" \
-d '{"name": "summarize_transcript", "arguments": {"transcript": "Your transcript here..."}}'
```
### MCP Client Libraries
This application works with any MCP-compliant client library:
- [MCP Python SDK](https://github.com/modelcontextprotocol/python-sdk)
- [MCP JavaScript SDK](https://github.com/modelcontextprotocol/js-sdk)
- [MCP Go SDK](https://github.com/modelcontextprotocol/go-sdk)
For more information about the MCP standard, visit the [official MCP specification](https://modelcontextprotocol.io/specification) and [GitHub repository](https://github.com/modelcontextprotocol/specification).
## Troubleshooting
- **"ModuleNotFoundError: No module named 'agents'"**: Make sure you're running commands from the project root directory
- **"uvicorn: command not found"**: Use `python3 -m uvicorn` instead of just `uvicorn`
- **API Key Issues**: Ensure `openai_key.txt` exists and contains a valid OpenAI API key
- **Port Conflicts**: Make sure ports 8000, 8001, and 8002 are available
- **MCP Discovery Issues**: Verify agents are running and accessible at their respective ports
## Development
### Code Formatting
To maintain consistent code style, use the provided formatting script:
```bash
python3 format_code.py
```
This will format all Python files with Black and HTML/Markdown files with Prettier.
### Manual Formatting
You can also format files individually:
```bash
# Format Python files
python3 -m black agents/ --line-length=88
# Format HTML and Markdown files
prettier --write index.html README.md
```
## File Structure
```
mcps/
├── agents/
│ ├── __init__.py
│ ├── summarizer_agent.py # MCP-compliant summarizer
│ ├── task_extractor_agent.py # MCP-compliant task extractor
│ ├── super_agent.py # Web UI coordinator & orchestrator
│ ├── models.py
│ ├── config.py
│ └── utils.py
├── docs/
│ └── images/
│ ├── welcome-page.png
│ ├── summarize-flow.png
│ ├── key-highlights-flow.png
│ └── task-extraction-flow.png
├── index.html
├── requirements.txt
├── README.md
├── format_code.py
└── openai_key.txt (create this file)
```