flow

This Flow MCP Server provides a standardized interface for AI assistants to interact with the Flow blockchain. It exposes tools to check balances (Flow and fungible tokens), resolve Flow domains (.find and .fn), execute Cadence scripts, submit and monitor transactions, and retrieve account information. When running in stdio mode, it integrates directly with AI tool configurations; when running as an HTTP server, it exposes endpoints for real-time updates, tool invocations, and health checks. You can select the Flow network (mainnet, testnet, or emulator) and point the server at a specific Flow access node via environment variables or command-line options. Available tools include: get_balance, get_token_balance, get_account, resolve_domain, execute_script, send_transaction, and get_transaction. These tools enable reading blockchain data, performing write operations, and monitoring on-chain activity.

To integrate with an AI assistant, run the server in stdio mode and configure the assistant’s tool settings to invoke the provided commands. For HTTP deployments, clients can POST to /messages with a tool_request payload to perform actions like balance inquiries or domain resolutions, and subscribe to updates via /sse for real-time feedback.

Prerequisites:

• Node.js (LTS) and npm installed on your system
• Git for cloning repositories (optional if using npx)

Local installation (for development):

1. git clone https://github.com/lmcmz/flow-mcp-server.git
2. cd flow-mcp-server
3. npm install
4. npm run build
5. npm start

Running with npx (recommended):

1. Ensure Node.js and npm are installed
2. Run the server in stdio mode (for AI assistants): npx -y @outblock/flow-mcp-server --stdio
3. To serve via HTTP on a port (e.g., 3000): npx -y @outblock/flow-mcp-server --port 3000

Environment variables (optional):

• FLOW_NETWORK: mainnet, testnet, or emulator
• FLOW_ACCESS_NODE: custom Flow access node URL
• PORT: HTTP port to listen on (used when not using stdio mode)
• LOG_LEVEL: debug, info, warn, or error

Tips and common considerations:

• If using HTTP mode, ensure the port you choose is open in your firewall and not in use by another service.
• FLOW_NETWORK and FLOW_ACCESS_NODE let you override the default Flow configuration; set them to point to your desired environment.
• The server supports real-time updates via Server-Sent Events at /sse when running in HTTP mode.
• For AI integrations, using stdio mode simplifies integration with tools configurations; the provided example demonstrates how to wire the tool into Claude-style tool configurations.
• If you encounter issues with formatting arguments (as hinted by version notes in some releases), ensure you are using compatible Node/npm versions and that you are invoking the correct binary by name (@outblock/flow-mcp-server).
• The tool set includes balance checks, token balances, domain resolution, script execution, transactions, and account lookups; be mindful of network latency when querying on mainnet.