# Legal MCP Server

A Model Context Protocol (MCP) server for legal document search and retrieval, powered by Azure Cosmos DB and Voyage AI embeddings.

## Overview

This MCP server exposes 12 search and retrieval tools for legal documents, providing semantic search, keyword search, and metadata-based queries over a corpus of vectorized legal documents.

## Features

- **Semantic Search**: Vector search over 3,275+ document chunks using Voyage AI voyage-law-2 embeddings
- **Keyword Search**: BM25-based full-text search for precise term matching
- **Bates Number Lookup**: Direct document retrieval by filing number
- **Claim-Specific Queries**: Filter documents by claim status and party
- **Document Mapping**: Cross-reference documents across the corpus
- **Statutory Citation Search**: Find legal precedents and statutory references
- **Compression Support**: Retrieve facts (compressed) or raw text for efficiency

## Architecture

### Azure Integration

- **Cosmos DB**: Stores 3,275 document chunks and 2,772 embeddings
- **Azure Key Vault**: Manages API keys and secrets (Voyage AI, etc.)
- **Azure Container Apps**: Hosts the MCP server

### Data Model

Documents are chunked and vectorized using Voyage AI's legal-specialized model (1024 dimensions).

See [SCHEMA.md](SCHEMA.md) for complete database schema documentation.

## Setup

### Requirements

- Python 3.10+
- Azure credentials (for Cosmos DB and Key Vault access)
- Voyage AI API key

### Installation

```bash
pip install -r requirements.txt
```

### Environment

Set the following environment variables:

- `COSMOS_DB_ENDPOINT`: Cosmos DB endpoint URL
- `COSMOS_DB_KEY`: Cosmos DB read key
- `VOYAGE_API_KEY`: Voyage AI API key
- `AZURE_KEY_VAULT_URL`: Key Vault URL

Or store them in Azure Key Vault (recommended for production).

## Usage

Start the MCP server:

```bash
python legal_mcp.py
```

The server exposes tools via the Model Context Protocol for use by Claude or other MCP clients.

## Tools

The server provides the following tools:

1. `semantic_search` - Search documents by meaning
2. `keyword_search` - Full-text search by keywords
3. `search_by_bates` - Find documents by Bates number
4. `search_by_header` - Find documents by section headers
5. `search_by_claim` - Find documents related to specific claims
6. `map_documents` - Cross-reference related documents
7. `search_citations` - Find statutory and case citations
8. And more...

## Deployment

The server is deployed as an Azure Container App in the personal Azure account.

**Status**: Active  
**Image**: `legalmcpacr.azurecr.io/legal-mcp:joinfix`  
**Endpoint**: `legal-mcp-server.purplesky-9ea73a76.eastus.azurecontainerapps.io`

## License

Private repository for legal case support.