Skip to content

Introducing the Aave MCP Server

Aave Labs

Aave Labs

The official Aave MCP server is live at mcp.aave.com. Any AI assistant that speaks the Model Context Protocol can now read live Aave data and build Aave transactions through one connection, covering both V3 and V4.

Ask an assistant for a wallet’s health factor, the best stablecoin yield, or a prepared supply of 500 USDC, and it calls Aave directly rather than working from training data or a third-party wrapper.

More and more people are reaching Aave through an assistant rather than a front end, and until now those assistants worked from training data or from third-party wrappers. Aave Labs built this MCP server so developers can connect an assistant to Aave in one line and trust what comes back, without maintaining their own bindings, learning two SDKs, or reimplementing accounting that already exists.

User-Controlled Signing

Every transaction the server builds comes back unsigned. The prepare_ tools return a transaction request or EIP-712 typed data, and the user’s own wallet signs it. The server holds no keys and cannot move funds.

What the Server Does

Reads cover chains, markets, and reserves, with rates, caps, and risk parameters, alongside time-series for APY, protocol deposits and borrows, and a wallet’s health factor over any window. Position queries return supplies, borrows, aggregate health, claimable rewards across every supported chain, and transaction history. On V4, positions carry per-position health, and Hub-level tools expose liquidity and accounting at the Liquidity Hub.

Transaction building covers supply, borrow, withdraw, repay, collateral toggles, eMode, rewards claims, and liquidations. A preview_action tool simulates any of them and reports the resulting health factor first. Swaps run through quote, prepare, submit, and status tools. Governance tools search proposals, return quorum status, and list voters by voting power.

One Endpoint for V3 and V4

Most tools take a version argument of v3, v4, or all. A single connection answers questions about either protocol, and all fetches both concurrently for direct comparison.

The server is built on Aave Kit, so results carry the same math as the official Aave interfaces, including Hub and Spoke accounting and Risk Premium. Responses are projected down to the fields a model needs, taking reserve detail from roughly 17 KB to roughly 1 KB, and rejected inputs return the reason verbatim so an assistant corrects its arguments rather than retrying a failed call.

What You Can Build

While we have more features on the way, builders can accomplish quite a lot with the server’s featureset. For example, ask for the best USDC supply rate across Aave and the server compares every deployment, then prepares the deposit for your wallet to sign. Ask to borrow 1,000 USDC against your ETH and it shows the rate and projected health factor before you approve.

If you want to go further, tell an agent to repay your USDC debt whenever your health factor drops below 1.5, and it monitors the position, prepares the repayment when the threshold hits, and waits for your signature. That same loop, read a position, simulate an action, prepare a transaction, is enough to build your own liquidation protection agent without writing a custom Aave integration.

Get Started

To get started, add https://mcp.aave.com as a custom connector in Claude or ChatGPT, or run the following in Claude Code:

claude mcp add --transport http aave https://mcp.aave.com

Cursor, VS Code, Windsurf, Codex CLI, and any other client supporting remote streamable HTTP connect with the URL alone. You can also check for Aave in the available connectors and plugins in the AI of your choice.

Review the skills here: github.com/aave/skills (and feel free to create new ones!)

Read the documentation here: aave.com/docs/mcp


Need help or want to learn more?

Share your questions or feedback and we'll get back to you.