Content
# Voice-to-Email-Outreach
A tool that converts voice notes into extracted insights - mathces them with a csv,ans sends professional emails to relevant recuriters.
hard coded as convex mcp(functionality soon)
## 🎯 How It Works
### 🎤 Voice → Text (ElevenLabs)
- **Input**: Audio file (M4A, MP3, etc.)
- **Process**: ElevenLabs Speech-to-Text API transcribes the audio
- **Output**: Raw text transcript
### 🤖 AgentMail
- **Input**: Raw transcript from voice -> extracts role and location -> matches with emails in csv file -> sends email to all ids using AgentMail
- **Process**: OpenAI GPT-4o-mini generates a professional email
- **Rules Applied**:
- Professional tone
- Slight storytelling elements
- No emojis (clean business style)
### 🔧 MCP → VibeFlow Convex MCP
- **Protocol**: Model Context Protocol (MCP)
- **Server**: Custom MCP server exposes the tool
- **Client**: Any MCP-compatible AI agent can call the tool
- **Result**: Structured JSON with transcript and LinkedIn post
## 📁 Project Structure
```
voice-linkedin/
├── agent.ts # MCP client that calls the tool
├── mcp-server.ts # MCP server with voice-to-LinkedIn tool
├── .env # API keys (ElevenLabs + OpenAI)
├── package.json # Dependencies
├── tsconfig.json # TypeScript configuration
└── Recording.m4a # Sample audio file for testing
```
## 🚀 Setup
### 1. Install Dependencies
```bash
npm install
```
### 2. Configure API Keys
Create `.env` file with your API keys:
```env
ELEVENLABS_API_KEY=your_elevenlabs_api_key_here
OPENAI_API_KEY=your_openai_api_key_here
```
### 3. Start the MCP Server
```bash
npx tsx mcp-server.ts
```
### 4. Use the Tool
```bash
npx tsx agent.ts
```
## 🛠️ Usage Example
### Input
- Audio file: `./Recording.m4a`
## 🔧 MCP Tool Details
### Tool Name
`voiceToEmail
### Parameters
- `filePath` (string): Path to the audio file to transcribe
### Returns
- `transcript`: Raw text from voice transcription
## 📋 Dependencies
- **@modelcontextprotocol/sdk**: MCP client/server implementation
- **axios**: HTTP client for API calls
- **form-data**: Multipart form data for file uploads
- **dotenv**: Environment variable management
- **zod**: Schema validation
## 🎯 Key Features
- **Voice-to-Text**: High-quality transcription using ElevenLabs
- **AI Content Generation**: Professional LinkedIn posts with GPT-4o-mini
- **MCP Protocol**: Compatible with any MCP-enabled AI agent
- **Error Handling**: Comprehensive error reporting and debugging
- **TypeScript**: Full type safety and IntelliSense support
## 🔍 Technical Details
### MCP Server Architecture
- Uses `McpServer` class from MCP SDK
- Registers `voiceToLinkedInPost` tool with proper schema
- Handles file validation and API error responses
- Returns structured JSON responses
### API Integration
- **ElevenLabs**: Speech-to-text with `scribe_v1` model
- **OpenAI**: Content generation with `gpt-4o-mini` model
- **Error Handling**: Detailed API error logging and response formatting
## 🐛 Troubleshooting
### Common Issues
1. **422 Error from ElevenLabs**
- Ensure `model_id` is set to `scribe_v1`
- Check API key is valid and has credits
2. **File Not Found**
- Verify audio file path exists
- Use relative paths like `./Recording.m4a`
3. **API Key Issues**
- Check `.env` file is in project root
- Verify API keys are correctly copied
### Debug Mode
Run with debug output:
```bash
npx tsx agent.ts
# Look for console.error messages in MCP server output
```
## 📄 License
MIT License - feel free to use this in your projects!
