https://github.com/sendaifun/devrel-mcp

Last synced: 2 months ago
JSON representation

Host: GitHub
URL: https://github.com/sendaifun/devrel-mcp
Owner: sendaifun
Created: 2025-06-30T21:03:29.000Z (3 months ago)
Default Branch: main
Last Pushed: 2025-07-03T19:56:14.000Z (3 months ago)
Last Synced: 2025-07-03T20:27:51.771Z (3 months ago)
Language: TypeScript
Size: 124 KB
Stars: 0
Watchers: 0
Forks: 0
Open Issues: 0
Metadata Files:
- Readme: README.md

Awesome Lists containing this project

README

# Jito Docs Search MCP

A powerful documentation search service using RAG (Retrieval Augmented Generation) built on top of the MCP infrastructure.

## Overview

This MCP provides semantic search capabilities for Jito's documentation using:
- Qdrant as the vector database
- OpenAI embeddings for semantic understanding
- Jina AI's cloud reranking service for improved result relevance
- Automatic indexing through GitHub workflows
- Real-time search via MCP endpoints

## Architecture

## Getting Started

### Prerequisites

- Node.js >= 18
- pnpm
- Docker (for local Qdrant)
- OpenAI API key
- GitHub fine-grained access token

### Installation

```bash
# Clone the repository
git clone https://github.com/your-org/jito-mcp
cd jito-mcp

# Install dependencies
pnpm install

# Setup environment variables
cp .env.example .env
# Edit .env with your credentials
```

### Local Development

1. Start Qdrant:
```bash
docker-compose up -d qdrant
```

2. Run the development server:
```bash
pnpm dev
```

3. Test the search endpoint:
```bash
curl -X POST http://localhost:3000/api/search \
-H "Content-Type: application/json" \
-d '{"query": "How to stake SOL?"}'
```

## Configuration

### GitHub Token Setup

1. Go to GitHub Settings -> Developer Settings -> Personal Access Tokens -> Fine-grained tokens
2. Create new token:
- Name: "Jito Docs Indexer"
- Repository access: "Only select repositories"
- Select both:
- Your MCP repository
- jito-foundation/jito-omnidocs
- Permissions:
- Repository permissions:
- Contents: Read
- Metadata: Read

### Jina AI API Key Setup

1. Go to [Jina AI](https://jina.ai/) and create an account
2. Navigate to the API dashboard
3. Generate a new API key
4. Add the key to your environment variables as `JINA_API_KEY`

**Benefits of Jina AI Cloud Reranking:**
- Eliminates the need for local model hosting (saves ~2GB RAM)
- Uses state-of-the-art `jina-reranker-v1-turbo-en` model
- Faster inference with optimized cloud infrastructure
- Automatic model updates and improvements
- Pay-per-use pricing with generous free tier

### Environment Variables

```env
QDRANT_URL=http://localhost:6333
QDRANT_API_KEY=your-qdrant-api-key
OPENAI_API_KEY=your-openai-api-key
JINA_API_KEY=your-jina-api-key
TOKEN_GITHUB=your-fine-grained-token
```

### Qdrant Collection Schema

```typescript
interface DocChunk {
id: string;
content: string;
metadata: {
path: string;
section: string;
title: string;
lastUpdated: string;
};
embedding: number[];
}
```

## Indexing Documentation

### Automatic Indexing (GitHub Workflow)

Documentation is automatically indexed when:
- Every 6 hours via scheduled workflow
- Manually triggered through GitHub Actions
- Webhook triggered from docs repository

To manually trigger reindexing:
```bash
pnpm run index-docs
```

### Manual Indexing

For local development or one-time indexing:

```bash
# Index all documentation
pnpm run index-docs

# Index specific section
pnpm run index-docs --section jitosol
```

## Usage

### MCP Integration

```typescript
import { mcpClient } from '@jito-labs/mcp-core';

const results = await mcpClient.jito_docs_search({
query: "How do I stake SOL for JitoSOL?",
section: "jitosol", // optional
limit: 5 // optional
});
```

### Response Format

```typescript
{
results: [
{
content: string;
score: number;
metadata: {
path: string;
section: string;
title: string;
lastUpdated: string;
}
}
]
}
```

## Development

### Project Structure

```
jito-mcp/
├── packages/
│ ├── docs-search/ # Main MCP package
│ │ ├── src/
│ │ │ ├── index.ts # MCP entry point
│ │ │ ├── search.ts # Search implementation
│ │ │ ├── indexer/ # Document indexing
│ │ │ ├── db/ # Database operations
│ │ │ └── utils/ # Utilities
│ │ ├── tests/
│ │ └── package.json
│ └── shared/ # Shared utilities
├── scripts/
├── docker/
└── package.json
```

### Adding New Features

1. Create a feature branch
2. Implement changes
3. Add tests
4. Submit PR

### Running Tests

```bash
# Run all tests
pnpm test

# Run specific test
pnpm test search
```

## Monitoring & Maintenance

### Health Checks

Monitor the service health at:
- `/health` - Service health
- `/metrics` - Prometheus metrics

### Common Issues

1. **Slow Search Response**
- Check Qdrant connection
- Verify embedding generation

2. **Failed Indexing**
- Check GitHub webhook logs
- Verify OpenAI API access
- Review document processing logs

## Contributing

1. Fork the repository
2. Create your feature branch
3. Commit your changes
4. Push to the branch
5. Create a new Pull Request

## License

[Add your license here]

ecosyste.ms

Data

Tools

Indexes

Applications

Experiments

Awesome

https://github.com/sendaifun/devrel-mcp

Awesome Lists containing this project

README