sui-butler

Sui Butler

Move to https://github.com/tamago-labs/sui-mcp

Sui Butler is a Model Context Protocol (MCP) server implementation for the Sui blockchain ecosystem that bridges AI intelligence for simplified development and more.

• YouTube Demo
• Website
• Presentation
• NPM Registry
• Backend Repo

Components

The system is composed of two subsystems:

• Sui Butler Client – (This repository) A Node.js TypeScript library designed to run inside MCP-compatible AI model clients such as Claude Desktop. It enables AI agents to interact with the Sui blockchain.
• Sui Butler Backend – The backend system built using the AWS Serverless Stack. It includes backend services and a dashboard for issuing access keys and managing transactions in zkLogin mode.

Features

• 30+ MCP tools covering account management, smart contract development, staking, token operations and market data
• Token swaps on Mainnet via the Cetus DEX Aggregator
• Pyth price oracle integration for real-time market data
• Sui CLI integration for smart contract development and testing
• Fully non-custodial, enables transactions using zkLogin wallets from the AI chat interface

Using with Claude Desktop

There are two mode available: zkLogin (recommended for most new users) and Private Key (for advanced users).

Private Key Mode

In Private Key mode, all operations (including transfers and other write operations) will be executed automatically without requiring additional approval.

1. Install Claude Desktop if you haven't already
2. Open Claude Desktop settings
3. Add the Sui MCP client to your configuration:

{
  "mcpServers": {
    "sui-butler": {
      "command": "npx",
      "args": [
        "-y",
        "sui-butler",
        "--sui_private_key=YOUR_PRIVATE_KEY", 
        "--sui_network=mainnet"
      ],
      "disabled": false
    }
  }
}

Private Key mode is recommended for advanced users who can securely manage their private keys. The MCP client handles transactions locally without exposing any data to external servers.

zkLogin Mode

With zkLogin authentication, read operations (balance checks, quotes) work immediately, but write operations (transfers, swaps) require approval in the dashboard.

1. Install Claude Desktop if you haven't already
2. Open Claude Desktop settings
3. Add the Sui MCP client to your configuration:

{
  "mcpServers": {
    "sui-butler": {
      "command": "npx",
      "args": [
        "-y",
        "sui-butler",
        "--sui_access_key=YOUR_ACCESS_KEY", 
        "--sui_network=mainnet"
      ],
      "disabled": false
    }
  }
}

The access key can be obtained from the dashboard. After logging in, a unique access key will be generated for each user.

Use Cases

1. DeFi Portfolio Management

Butler connects to Pyth price oracles and external sources to help you:

• Monitor real-time cryptocurrency prices across multiple assets
• Compare prices across different platforms for optimal trading opportunities
• Execute token swaps via Cetus Aggregator

Example:

2. Smart Contract Development & Testing Assistance

Butler integrates with the Sui CLI to help developers:

• Analyze existing Move code and suggest improvements
• Generate comprehensive test cases for smart contracts
• Publish and upgrade packages directly through AI conversation

Example:

3. Protocol Governance & Parameter Management

Butler assists DeFi protocol managers with:

• Checking external sources to determine optimal parameters based on current market conditions
• For example, in collateralization protocols, Butler can analyze asset prices to suggest better collateral ratio settings for smart contracts
• Then propose new governance parameters through AI conversations

Example:

Background

Today, when building AI applications—especially those focused on crypto, we often rely on agent kits based on Langchain. These kits tightly couple the AI model and components, requiring frequent updates, otherwise, the application risks becoming non-functional within a few months or even weeks.

Model Context Protocol (MCP), introduced by Claude AI in late 2024, has quickly become popular today. It solves this issue by integrating directly with AI interfaces, allowing users to easily switch to the latest models and interact with Web3 through standardized tools.

Available Tools

Wallet Operations

Token Transfers & DeFi

Staking Operations

Token Management

SNS Domain Services

Sui CLI Integration

Price Data (Pyth)

zkLogin Transaction Flow

When a user operates in zkLogin mode using an MCP-compatible AI client:

1. The client pushes a transaction request to the backend.

2. The transaction is stored in the database with a pending status.

3. The user can visit the dashboard to manually approve the transaction using their zkLogin-authenticated session.

Troubleshooting

If you're using Ubuntu or another Linux environment with NVM, you'll need to manually configure the path. Follow these steps:

1. Install the Sui Butler under your current NVM-managed Node.js version.

npm install -g sui-butler

2. Due to how NVM installs libraries, you may need to use absolute paths in your config. Replace the example values below with your actual username and Node version:

{
  "mcpServers": {
    "sui-mcp": {
      "command": "/home/YOUR_NAME/.nvm/versions/node/YOUR_NODE_VERSION/bin/node",
      "args": [
        "/home/YOUR_NAME/.nvm/versions/node/YOUR_NODE_VERSION/bin/sui-butler",
        "--sui_access_key=YOUR_ACCESS_KEY",
        "--sui_network=mainnet"
      ]
    }
  }
}

3. Restart Claude Desktop and it should work now.

Work with Local Files

When working with local files especially when using Sui CLI tools for smart contract development to create, build, and test a Move package on your machine—you’ll need to import an additional MCP server library of filesystem made by Claude team. Use with:

"filesystem": {
  "command": "npx",
  "args": [
    "-y",
    "@modelcontextprotocol/server-filesystem",
    "${workspaceFolder}"
  ],
  "disabled": false
}

workspaceFolder refers to your working directory. You can provide more than one argument. Subfolders or specific files can then be referenced in your AI prompt.

If you're using Linux and encounter issues during setup, please refer to the troubleshooting section.

License

This project is licensed under the MIT License.