u2-mcp

Connect Claude to your Universe/UniData database using natural language.

u2-mcp is an MCP server that lets AI assistants like Claude query and interact with Rocket Universe and UniData MultiValue databases. Ask questions in plain English and get real answers from your data.

The first MCP server for the Pick/MultiValue database family.

What Can You Do With This?

Instead of writing RetrieVe queries, just ask Claude:

How It Works

┌─────────────────┐     ┌─────────────────┐     ┌─────────────────┐
│  Claude Desktop │◄───►│     u2-mcp      │◄───►│ Universe/UniData│
│  (You ask here) │     │  (Translates)   │     │   (Your data)   │
└─────────────────┘     └─────────────────┘     └─────────────────┘

1. You ask Claude a question about your database
2. Claude calls u2-mcp with the appropriate tool
3. u2-mcp queries your Universe/UniData server
4. Results come back through Claude in a readable format

New to MCP? See What is MCP? for a complete explanation.

Documentation

Features

• Connection Management - Connect to Universe/UniData servers with connection pooling and SSL support
• File Operations - Read, write, and delete records while preserving MultiValue semantics
• Query Execution - Run RetrieVe/UniQuery statements with safety controls
• Dictionary Access - Explore file structures and field definitions
• BASIC Subroutine Calls - Execute cataloged BASIC programs
• Transaction Support - Full transaction management with commit/rollback
• AI-Optimized - Built-in query examples and syntax help for better LLM interactions

Installation

pip install u2-mcp

Or using uvx for isolated execution:

uvx u2-mcp

Quick Start

Prerequisites

• Python 3.10+
• Access to a Rocket Universe or UniData server
• The uopy package (installed automatically as a dependency)

Configuration

Set environment variables for your database connection:

export U2_HOST=server.example.com
export U2_USER=username
export U2_PASSWORD=password
export U2_ACCOUNT=MYACCOUNT
export U2_SERVICE=uvcs  # or 'udcs' for UniData

Claude Desktop Configuration

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "u2": {
      "command": "uvx",
      "args": ["u2-mcp"],
      "env": {
        "U2_HOST": "server.example.com",
        "U2_USER": "user",
        "U2_PASSWORD": "password",
        "U2_ACCOUNT": "MYACCOUNT"
      }
    }
  }
}

Usage Examples

Once connected, you can use natural language to interact with your Universe database:

• "List all customers in California"
• "Read record CUST001 from the CUSTOMERS file"
• "Show me the dictionary for the ORDERS file"
• "Count all open invoices over $1000"

Available Tools

Connection Management

• connect - Establish connection to Universe/UniData server
• disconnect - Close connections
• list_connections - Show active connections

File Operations

• list_files - List available files in the account
• read_record - Read a single record by ID
• read_records - Read multiple records
• write_record - Write/update a record
• delete_record - Delete a record
• get_file_info - Get file statistics

Query Execution

• execute_query - Run RetrieVe/UniQuery statements
• execute_tcl - Run TCL/ECL commands
• get_select_list - Execute SELECT and return record IDs

Dictionary Access

• list_dictionary - List dictionary items
• get_field_definition - Get field details
• describe_file - High-level file description

Advanced Features

• call_subroutine - Call BASIC subroutines
• begin_transaction / commit_transaction / rollback_transaction - Transaction management

Knowledge Persistence

• save_knowledge - Save learned information about the database
• list_knowledge - List all saved knowledge topics
• get_knowledge_topic - Retrieve specific topic
• search_knowledge - Search across saved knowledge
• delete_knowledge - Remove a knowledge topic

Configuration Options

Knowledge Persistence

The MCP server includes a knowledge persistence feature that allows Claude to save and recall learned information about your database across sessions. This eliminates repetitive discovery work and speeds up future interactions.

How It Works

When Claude discovers useful information about your database (file purposes, field meanings, working queries), it can save this knowledge using the save_knowledge tool. This information is stored in a markdown file and automatically available in future conversations via the u2://knowledge resource.

What Gets Saved

• File descriptions - What each file contains (e.g., "AR-CUST is the customer master file")
• Field mappings - Which fields contain what data (e.g., "Field 1 is customer name")
• Query patterns - Queries that produced good results
• Data formats - Date formats, code meanings, conversion notes
• Relationships - How files relate to each other

Storage Location

Knowledge is stored in ~/.u2-mcp/knowledge.md by default. You can customize this location using the U2_KNOWLEDGE_PATH environment variable:

{
  "env": {
    "U2_KNOWLEDGE_PATH": "/path/to/custom/knowledge.md"
  }
}

Example Usage

After Claude discovers that AR-CUST contains customer records:

Claude: I found that AR-CUST is the customer master file. Let me save this for future reference.
[Calls save_knowledge("AR-CUST file", "Customer master file. Key is customer number. Field 1=name, Field 2=address...")]

In the next conversation, Claude will already know this and won't need to rediscover it.

Centralized Deployment (HTTP/SSE Mode)

For team environments, you can run u2-mcp as a centralized HTTP server instead of local stdio mode. This allows:

• Single database connection for the whole team
• Shared knowledge base
• Centralized credential management
• Access from any MCP-compatible client

Running the HTTP Server

# Basic HTTP server on default port 8080
u2-mcp --http

# Custom host and port
u2-mcp --http --host 0.0.0.0 --port 3000

# Or using environment variables
export U2_HTTP_HOST=0.0.0.0
export U2_HTTP_PORT=3000
u2-mcp --http

Docker Deployment

FROM python:3.12-slim

RUN pip install u2-mcp

ENV U2_HOST=your-universe-server
ENV U2_USER=username
ENV U2_PASSWORD=password
ENV U2_ACCOUNT=MYACCOUNT
ENV U2_HTTP_PORT=8080

EXPOSE 8080

CMD ["u2-mcp", "--http"]

docker build -t u2-mcp-server .
docker run -p 8080:8080 u2-mcp-server

HTTP Configuration Options

Connecting Clients

The HTTP server exposes an SSE (Server-Sent Events) endpoint at /sse. MCP clients that support remote servers can connect to:

http://your-server:8080/sse

Note: As of late 2024, Claude Desktop only supports local stdio MCP servers. HTTP/SSE mode is for:

• Custom MCP clients
• Future Claude Desktop versions with remote server support
• Integration with other AI tools that support MCP over HTTP

Claude.ai Integration (Streamable HTTP + OAuth)

Deploy u2-mcp as a Custom Connector in Claude.ai with enterprise authentication via your identity provider (Duo, Auth0, Okta, etc.).

Quick Setup

1. Deploy u2-mcp with OAuth enabled:

u2-mcp --streamable-http --host 0.0.0.0 --port 8080

2. Configure OAuth in your .env:

U2_AUTH_ENABLED=true
U2_AUTH_ISSUER_URL=https://u2-mcp.example.com
U2_IDP_PROVIDER=duo
U2_IDP_DISCOVERY_URL=https://sso-xxx.sso.duosecurity.com/oidc/CLIENT_ID/.well-known/openid-configuration
U2_IDP_CLIENT_ID=your_client_id
U2_IDP_CLIENT_SECRET=your_client_secret

3. Add to Claude.ai: Go to Settings → Connectors → Add Custom Connector → Enter your URL

Users authenticate through your IdP and can then query your database directly from Claude.ai.

Supported Identity Providers

• Cisco Duo - Enterprise MFA and SSO
• Auth0 - Flexible identity platform
• Any OIDC Provider - Okta, Azure AD, Google Workspace, Keycloak, etc.

See the OAuth Guide for detailed setup instructions.

Security

• Credentials are never logged
• Configurable command blocklist prevents dangerous operations
• Optional read-only mode for safe exploration
• Honors Universe's built-in security model

See SECURITY.md for reporting security vulnerabilities.

Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

Documentation

Getting Started

• Quickstart Guide - Get running in 10 minutes
• What is MCP? - Understanding MCP and how u2-mcp works
• Installation Guide - Detailed installation instructions

Reference

• Configuration Reference - All environment variables
• Tools Reference - Complete tool documentation
• Usage Examples - Common usage patterns
• OAuth & Claude.ai Integration - Deploy as a Claude.ai Custom Connector

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

Acknowledgments

• Rocket Software for the uopy package
• Anthropic for the Model Context Protocol specification
• The MultiValue database community