Aave data and transactions
Aave helps users explore live Aave V3 and V4 markets, review wallet positions and DAO governance, simulate lending actions, and prepare non-custodial transactions or signed-order workflows through your AI.
Example path → ·
https://mcp.aave.comAave MCP: one endpoint serves both Aave V3 and V4. Pass a 'version' argument ('v3', 'v4', or 'all') to most tools; a tool without that argument states its applicability in the first words of its description ('v4 only', 'v3 only', or protocol-agnostic). Non-custodial: read and prepare_* tools return data or an unsigned transaction / EIP-712 typed data for the user's own wallet to sign; this server never holds keys or submits (only and relay what the user already signed). Workflow: discover (, or on a cold turn) -> inspect (, ) -> simulate (, on either version) -> build () -> the user's wallet signs. Do not skip the simulation before a borrow or a withdraw. Every successful result is the same envelope, { data, next_actions? }, returned both as JSON text and as structuredContent; read the payload from 'data' and follow 'next_actions'. Inside 'data', six reads nest their payload under 'v4' and 'v3' - , , , , , - including when you asked for a single version, so on a version:'v4' call the answer is at 'data.v4' and 'data' itself holds no fields. Every other tool returns its payload on 'data' directly, and the prepare_* tools included, even though they take the same 'version' argument. A result may also carry 'warnings', each with a level: 'error' means the action cannot succeed as built, so stop and fix it or tell the user rather than preparing or signing past it; 'warning' means it will succeed but the user should be told; 'info' is context. Every USD figure is Aave's oracle price for the asset, read from Chainlink price feeds and named as 'priceSource' on a reserve read: when you quote a price or a USD value, attribute it to Chainlink rather than presenting it as a market quote. Token amounts are in main units (e.g. '10.5'), never wei. To build an action you need a reserveId (v4) or market+token+chainId (v3) from in this session - these are deployment-specific, so an Aave address or id you already recognise belongs to a different deployment and is rejected; call before the first prepare rather than after it refuses; narrow that call with 'symbols' (e.g. ['USDC']) instead of listing every reserve. Reads cover every supported chain by default and say which under 'chainsCovered', so do not loop over chains by hand; where a read is narrower it also returns 'chainsNotCovered', which is the only case a sweep is needed. A chain under 'chainsNotServed' is one this API holds no market on: an empty result there is an absence of data, not an answer about the chain, and asking again returns the same nothing - flags those chains too. Rank on a rate that can actually be entered, not on APY alone: carries the liquidity left and flags a reserve that is frozen, paused or at its cap. Health factor and aggregate values come from v4's cross-chain summary; v3 is per-market. Token swaps are protocol-agnostic (they run on the v4 backend). Aave DAO governance tools (, , , , ) take no 'version' either: Governance V3 is the DAO's own contract generation, unrelated to the v3 / v4 markets. Leave an optional argument out of the call entirely when you are not using it; do not send an empty string, a zero or a near-zero number to stand in for 'unset'. A tool error that says the input was rejected will fail again unchanged: fix the arguments instead of retrying. When an Aave conversation opens with no specific request - the user is asking what this can do, or just says hello - call and answer from it rather than guessing at the tool list. Savings GHO (sGHO) is v3 and Ethereum only, and a withdrawal there is denominated in vault shares rather than in GHO: read the 'gho' guide topic before using or . When unsure how something works, call (topics: overview, v4, v3, positions, health-factor, risks, ids, amounts, prices, signing, swaps, gho, safety-module, portfolio, governance, rewards, tools, docs).
5 in the ChatGPT listing