Aave Labs just made it a lot easier for AI agents to talk to its lending protocols. The team launched a Model Context Protocol (MCP) server that gives AI applications a single, standardized endpoint to pull live data from both Aave V3 and V4, replacing the patchwork of static datasets and third-party wrappers that developers previously had to cobble together.
The server, accessible at mcp.aave.com, connects to Aave V3 deployments across 21 different blockchains and to Aave V4 on Ethereum and Avalanche. Think of it as a universal translator between AI models and Aave’s on-chain infrastructure.
What the MCP server actually does
Model Context Protocol, or MCP, is a standardized way for AI applications to access external data and tools in real time. Aave’s implementation offers approximately 40 tools that cover everything from market data retrieval to transaction preparation.
Users and AI agents can check wallet positions, examine health factors (the metric that determines how close a position is to liquidation), simulate potential actions before committing capital, and prepare unsigned transactions. That last part matters: the server is non-custodial by design, meaning it can assemble a transaction for you but never holds your keys or signs anything on your behalf.
For a concrete example: an AI portfolio manager could now query a user’s Aave positions across multiple chains, identify that a health factor on one position is trending dangerously low, simulate a partial repayment to see how it would improve the ratio, and prepare the exact transaction needed to execute it. All in one flow, all from one data source.
Why this matters for DeFi’s AI race
Aave’s approach is notable for its scope. Supporting V3 across 21 chains means the MCP server covers the vast majority of Aave’s deployed capital. Adding V4 on Ethereum and Avalanche signals that the team views this integration layer as forward-looking, not just a convenience feature bolted onto legacy infrastructure.
The non-custodial architecture is a deliberate design choice that addresses one of the thorniest questions in the AI-agent space: who controls the keys? By limiting the server to unsigned transactions, Aave sidesteps the trust problem entirely. An AI agent can do everything up to the point of execution, but a human (or a separate, purpose-built signing module) still has to approve the final step.
The roughly 40 tools available at launch suggest Aave is thinking about this comprehensively rather than offering a minimal viable product. Market data, position management, risk simulation, and transaction preparation cover the core workflows that any AI-powered DeFi application would need.