Aave Labs has launched an official Model Context Protocol server that allows compatible AI assistants to query live protocol data and build transactions across Aave V3 and V4. The September 8 release creates a standardized interface between AI software and Aave while leaving transaction signing under the user’s control. The server is live as a remote endpoint and can be connected to MCP-compatible tools including ChatGPT, Claude, Cursor, VS Code and Codex CLI.
According to Aave’s official launch announcement, assistants can retrieve market rates, reserve parameters, wallet positions, health factors, rewards and historical protocol data before preparing actions such as supplying, borrowing, withdrawing or repaying assets. The MCP server gives AI systems direct access to current Aave information instead of forcing them to rely on training data or independently maintained integrations. It also supports governance queries, including proposal status, quorum and voting-power data.
AI Can Prepare Actions but Cannot Sign Them
The transaction boundary is deliberately restrictive. Aave’s prepare tools return unsigned transaction requests or EIP-712 typed data, while preview_action can simulate supply, borrow, withdraw or repay operations and calculate the resulting health factor before anything is executed. The MCP server holds no private keys and cannot independently move assets from a user’s wallet. Final authorization remains a separate wallet action.
That structure also applies to more automated workflows. Aave gives the example of an agent monitoring a borrower’s health factor and preparing a repayment if it falls below a chosen threshold. An AI assistant can monitor conditions, decide when its programmed trigger has been reached and construct the required transaction, but it still waits for the user’s signature before the repayment reaches the blockchain. This creates room for machine-assisted position management without giving the MCP server direct custody authority.
The server extends beyond lending actions. It can quote token swaps, prepare orders and track their settlement, while governance tools can search proposals and inspect voting activity. Two order-related tools can relay already signed orders or cancellations, but neither can create the user’s signature. The distinction is between coordinating execution and possessing execution authority, a security boundary that remains intact even as more of the surrounding workflow becomes machine-readable.
Aave Builds an Interface for Software Agents
Aave founder Stani Kulechov has framed the development as part of a broader shift toward AI agents allocating capital and managing financial positions. The current implementation, however, remains more constrained than fully autonomous fund management. Aave has exposed the data, simulation and transaction-construction layers to agents while keeping the decisive cryptographic approval outside the AI system. That makes the launch infrastructure for agent-assisted DeFi rather than a transfer of wallet control to autonomous software.
The MCP server is built on Aave Kit and exposes both V3 and V4 through a single connection, including V4 Liquidity Hub accounting and position-level health information. That unified interface reduces the integration work required for developers building conversational portfolio tools, risk monitors or transaction assistants around multiple versions of Aave. Aave has also published reusable skills covering workflows such as yield analysis, deleveraging, account activity and transaction confirmation.
The next meaningful indicator will be usage rather than the existence of the endpoint itself. Developer integrations, agent-generated transaction volume and adoption of automated monitoring workflows will show whether MCP becomes a meaningful access layer for Aave, while the continued requirement for wallet signatures will determine how far automation can advance without changing the protocol’s current custody model.