defi-mcp
An MCP server that gives AI agents direct, on-chain access to DeFi data on EVM chains — wallet balances, Aave v3 lending positions, and live token prices, with no centralized API in the middle.
README
defi-mcp
An MCP (Model Context Protocol) server that gives AI agents direct, on-chain access to DeFi data on EVM chains — wallet balances, Aave v3 lending positions, and live token prices, with no centralized API in the middle.
Status: Live and functional. 5 tools available today, more on the roadmap.
Why
Most "DeFi + AI" integrations route through a centralized API that can rate-limit, paywall, or simply disappear. defi-mcp talks directly to the chain (via your own RPC provider) and to first-party oracles (Chainlink), so the data your AI agent gets is exactly what's on-chain — verifiable, and not dependent on a third-party API staying online.
Tools
get_token_balance
Reads the ERC-20 balance of any wallet address for a supported token.
Supported tokens:
- Polygon:
USDC,USDC.e,USDT,WETH,WMATIC,WBTC,DAI,AAVE - Arbitrum:
USDC,USDC.e(bridged),USDT,WETH,WBTC,ARB
// get_token_balance(address="0x...", token_symbol="USDC", chain="polygon")
{
"address": "0x...",
"token": "USDC",
"chain": "polygon",
"balance": 268.21,
"raw_balance": "268210000",
"decimals": 6
}
get_aave_position
Reads a user's full Aave v3 lending/borrowing position: collateral, debt, available borrowing power, and health factor — straight from the Aave v3 Pool contract.
// get_aave_position(address="0x...", chain="polygon")
{
"address": "0x...",
"chain": "polygon",
"total_collateral_usd": 969.78,
"total_debt_usd": 288.16,
"available_borrows_usd": 484.86,
"liquidation_threshold_pct": 82.79,
"ltv_pct": 79.71,
"health_factor": 2.79,
"has_active_position": true
}
A health_factor below 1.0 means the position is at risk of liquidation. A null health factor means the user has no outstanding debt.
get_token_price
Reads the current USD price of a token directly from a Chainlink price feed — the same oracle DeFi protocols use internally, so the price is consistent with what get_aave_position reports.
Supported feeds:
- Polygon:
ETH/WETH,MATIC/WMATIC,WBTC,USDC,USDT - Arbitrum:
ETH/WETH,WBTC,ARB,USDC,USDT
// get_token_price(token_symbol="WBTC", chain="polygon")
{
"token": "WBTC",
"chain": "polygon",
"price_usd": 63835.77,
"decimals": 8,
"updated_at": 1782032816
}
simulate_swap
Simulates a Uniswap v3 swap on-chain via the QuoterV2 contract — no transaction submitted, no wallet needed. Tries all four fee tiers (0.01%, 0.05%, 0.3%, 1%) and returns the best available output amount.
Supported tokens:
- Polygon:
USDC,USDC.e,USDT,WETH,WMATIC,WBTC,DAI,AAVE - Arbitrum:
USDC,USDC.e(bridged),USDT,WETH,WBTC,ARB
// simulate_swap(token_in="WMATIC", token_out="USDC", amount_in=100.0, chain="polygon")
{
"chain": "polygon",
"token_in": "WMATIC",
"token_out": "USDC",
"amount_in": 100.0,
"amount_out": 7.990547,
"fee_tier_used_pct": 0.05,
"fee_tier_used_bps": 500,
"gas_estimate": 122770
}
If no pool exists for the pair at any fee tier, returns {"error": "No liquidity found for this pair on any fee tier"}.
hello
Simple connectivity check — confirms the MCP server is reachable and responding.
Supported chains
| Chain | get_token_balance |
get_aave_position |
get_token_price |
simulate_swap |
|---|---|---|---|---|
| Polygon | ✅ | ✅ | ✅ | ✅ |
| Arbitrum | ✅ | ✅ | ✅ | ✅ |
Installation
Requires Python 3.10+ and an RPC provider API key (e.g. Alchemy, free tier works).
pip install defi-mcp
Create a .env file in your working directory (see .env.example):
POLYGON_RPC_URL=https://polygon-mainnet.g.alchemy.com/v2/YOUR_API_KEY
ARBITRUM_RPC_URL=https://arb-mainnet.g.alchemy.com/v2/YOUR_API_KEY
Each chain only needs its RPC URL configured. Tools called with chain="arbitrum" will fail gracefully if ARBITRUM_RPC_URL is not set.
Usage with Claude Desktop
Add this to your Claude Desktop MCP config (claude_desktop_config.json):
{
"mcpServers": {
"defi-mcp": {
"command": "uvx",
"args": ["defi-mcp"],
"env": {
"POLYGON_RPC_URL": "https://polygon-mainnet.g.alchemy.com/v2/YOUR_API_KEY",
"ARBITRUM_RPC_URL": "https://arb-mainnet.g.alchemy.com/v2/YOUR_API_KEY"
}
}
}
}
Restart Claude Desktop, and the tools above will be available to the model.
Testing locally
You can test the server directly with the official MCP Inspector:
npx @modelcontextprotocol/inspector uvx defi-mcp
Roadmap
- [ ] Track Uniswap v3 LP positions and impermanent loss
- [ ] Monitor protocol TVLs
- [ ] Additional chains (Ethereum, Base)
- [ ]
defi-mcp-cloud— hosted tier with MEV-specific tools, caching, and higher rate limits
Architecture
Open-core model: this repository (MIT licensed) covers standard on-chain read tools. A separate defi-mcp-cloud will offer a hosted version with MEV-related tools, request caching, and managed RPC access for users who don't want to run their own infrastructure.
Resources
- defi-storage-cheatsheet — verified EVM storage slot layouts for Aave v3 and Uniswap v3, with
web3.pysnippets. Useful if you want to read on-chain state directly viaeth_getStorageAtinstead of callingviewfunctions — the lower-level approach behind some of what this server does under the hood.
Contributing
Issues and PRs welcome. This is an early-stage project — feedback on what tools would actually be useful to you is especially valuable.
License
MIT
<!-- mcp-name: io.github.matiosera3-ops/defi-mcp -->
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。