Lido MCP Server

Your AI agent can stake, unstake, vote, and monitor yields — without you writing a single contract call.

An MCP (Model Context Protocol) server that makes Lido staking operations natively callable by any AI agent. Say "Stake 1 ETH with Lido (dry run first)" and your agent handles the rest — against real Ethereum mainnet contracts, not mocks.

Lido.MCP.mp4

Features

• Staking: Stake ETH → stETH, request withdrawals, wrap/unwrap stETH ↔ wstETH
• Balances & Rewards: Query holdings, estimate staking rewards, protocol stats
• Governance: List Lido DAO proposals, cast votes via Aragon
• Vault Monitoring: Track Lido Earn vault health, yield benchmarks, alerts
• Dry Run Mode: Every write operation supports simulation before execution
• No Mocks: All data comes from live Ethereum mainnet contracts

Quick Start

Install and Build

git clone https://github.com/ghost-clio/lido-mcp.git
cd lido-mcp
npm install
npm run build

Connect to Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "lido": {
      "command": "node",
      "args": ["/path/to/lido-mcp/dist/index.js"],
      "env": {
        "ETH_RPC_URL": "https://eth.llamarpc.com",
        "ETH_PRIVATE_KEY": "0x..."
      }
    }
  }
}

Connect to Cursor

In Cursor settings → MCP Servers, add:

{
  "lido": {
    "command": "node",
    "args": ["/path/to/lido-mcp/dist/index.js"]
  }
}

Environment Variables

Variable	Required	Description
ETH_RPC_URL	No	Ethereum RPC endpoint (default: https://eth.llamarpc.com)
ETH_PRIVATE_KEY	For writes	Hex-encoded private key for staking/voting transactions

Read-only operations (balances, stats, proposals) work without a private key.

Try it

You: "How much stETH does vitalik.eth hold?"
Agent: [calls lido_balance] → 0.0 stETH, 0.0 wstETH, 4,291.7 ETH

You: "Stake 1 ETH with Lido, dry run first"
Agent: [calls lido_stake with dry_run=true] → Simulation passed. Gas: 82,341. Ready to execute.

Available Tools

Staking

Tool	Description	Params
lido_stake	Stake ETH → stETH	amount, dry_run
lido_unstake	Request withdrawal stETH → ETH	amount, dry_run
lido_wrap	Wrap stETH → wstETH	amount, dry_run
lido_unwrap	Unwrap wstETH → stETH	amount, dry_run

Queries

Tool	Description	Params
lido_balance	ETH/stETH/wstETH balances	address
lido_rewards	Estimated staking rewards	address, days
lido_stats	Protocol-wide statistics	—

Governance

Tool	Description	Params
lido_proposals	List DAO proposals	status, limit
lido_vote	Vote on a proposal	proposal_id, support, dry_run

Vault Monitor

Tool	Description	Params
vault_health	Vault status and allocations	vault
vault_alerts	Vault alerts and changes	vault, since
vault_benchmark	Yield vs stETH/Aave benchmarks	vault

Architecture

lido-mcp/
├── src/
│   ├── index.ts              # MCP server entry point (stdio transport)
│   ├── tools/
│   │   ├── staking.ts        # lido_stake, lido_unstake, lido_wrap, lido_unwrap
│   │   ├── balance.ts        # lido_balance, lido_rewards, lido_stats
│   │   ├── governance.ts     # lido_proposals, lido_vote
│   │   └── vault.ts          # vault_health, vault_alerts, vault_benchmark
│   ├── lib/
│   │   ├── lido.ts           # Lido contract interactions via viem
│   │   ├── vault-monitor.ts  # Earn vault monitoring + benchmarking
│   │   └── constants.ts      # Contract addresses, ABIs
│   └── types.ts              # TypeScript interfaces
├── lido.skill.md             # Agent skill file (teaches Lido mental model)
├── package.json
├── tsconfig.json
└── README.md

How It Works

1. The MCP server starts on stdio and registers 12 tools
2. Read operations use a public RPC (no API key needed) via viem's publicClient
3. Write operations require ETH_PRIVATE_KEY and use viem's walletClient
4. Dry runs simulate transactions via eth_call / estimateGas — no state changes
5. All amounts are in human-readable format (ETH, not wei) — conversion is handled internally

Contract Addresses (Ethereum Mainnet)

Contract	Address
stETH	0xae7ab96520DE3A18E5e111B5EaAb095312D7fE84
wstETH	0x7f39C581F595B53c5cb19bD0b3f8dA6c935E2Ca0
Withdrawal Queue	0x889edC2eDab5f40e902b864aD4d7AdE8E412F9B1
Aragon Voting	0x2e59A20f205bB85a89C53f1936454680651E618e

Examples

Check balances

> Use lido_balance for address 0xd8dA6BF26964aF9D7eEd9e03E53415D37aA96045

Simulate staking

> Stake 1 ETH with Lido (dry run first)

Monitor vault

> How is the Earn ETH vault doing? Compare it to stETH and Aave.

Check governance

> Show me the latest Lido DAO proposals

Tests

npm run build
node --test test/tools.test.cjs    # 23 tests

Suite	Tests	What it checks
Project structure	12	All source files exist
Tool registration	5	Correct tool count per module (4+3+2+3=12)
Tool names	1	All 12 tool names present in source
Dry run support	1	Write tools accept dry_run parameter
Contract addresses	2	stETH + wstETH addresses match Lido docs
Build output	1	dist/index.js exists after compilation
Skill file	1	lido.skill.md contains Lido mental model

FAQ

Why MCP over a REST API? MCP is a standard protocol that lets AI agents call tools natively — no HTTP client, no auth tokens, no parsing. The agent says "stake 1 ETH" and the MCP server handles serialization, RPC calls, and response formatting. It works with Claude Desktop, Cursor, and any MCP-compatible client out of the box.

Can this compose with other DeFi MCPs? Yes. Stake with this MCP, then pass the resulting wstETH address to an Aave MCP for lending. MCP tools are independent — agents orchestrate across them naturally.

Who's the target user? Anyone building AI agents that interact with Lido — DeFi automation, portfolio management bots, governance participation agents, or staking chatbots. The skill file (lido.skill.md) makes it accessible even to agents with no DeFi training.

Known Limitations

• Gas estimation: Dry runs simulate but don't report gas cost in the response. Agents can't yet make cost-aware staking decisions.
• Testnet: Only mainnet contracts are configured. Set ETH_RPC_URL to a Holesky RPC and update contract addresses in constants.ts for testnet use.
• No caching: Read operations hit the RPC every call. For high-frequency agents, add application-level caching.
• Vault scope: Monitors Lido Earn vaults only, not SimpleDVT or CSM modules.

Agent Skill File

The lido.skill.md file teaches AI agents the Lido mental model — rebasing mechanics, stETH vs wstETH tradeoffs, safe staking patterns, common pitfalls, and workflow examples. Point your agent to this file for best results.

Links

• Lido Documentation
• Lido Deployed Contracts
• MCP Specification
• Lido DAO Governance

Part of the Agent Stack

• AgentScope — On-chain policy enforcement for agent wallets (Safe module)
• Aegis — Autonomous treasury with pre-signing enforcement, smart DCA, yield hunting
• Ghost Protocol — Private reasoning (Venice.ai) + scoped execution

License

MIT