Architecture

How Moly works under the hood

Moly has three packages: the CLI (@moly-mcp/lido), a web dashboard on Vercel, and these docs.

Request Flow (CLI MCP Server)

AI Agent                  Moly CLI (stdio)            Ethereum
  │                            │                            │
  │── MCP tool call ──────────►│                            │
  │                            │── eth_call (viem) ─────────►│
  │                            │◄── balance data ───────────│
  │◄── tool result ────────────│                            │
  │                            │                            │

Components

Component	Purpose
CLI (npx @moly-mcp/lido)	28 tools via stdio MCP server + interactive terminal
OWS Vault	Encrypted wallet storage via Open Wallet Standard
/api/chat	AI chat — Vercel AI SDK + OpenRouter
/api/skill	Agent skill document (raw markdown)
/docs	Documentation (this site)
/dashboard	Chat dashboard UI

Stack

Next.js — app framework, deployed on Vercel
viem — raw Ethereum contract calls (no SDK dependencies)
@modelcontextprotocol/sdk — MCP server with stdio transport
Vercel AI SDK v6 — chat endpoint with tool calling
OpenRouter — LLM gateway for the dashboard chat

How Tools Work

All tools use raw viem contract calls against Lido's deployed contracts. Read tools call eth_call directly. Write tools in simulation mode return gas estimates and expected outputs. In live mode with a configured wallet (OWS or raw key), write tools broadcast real transactions.

Configuration (mode, network, wallet) is stored in ~/.moly/config.json and can be changed via the setup wizard or the update_settings tool mid-session.