AI Agents

Give your agent a wallet it can actually read.

Positions, balances and activity as structured tool responses over MCP, or as JSON over REST — so the agent reasons on decoded, USD-valued data instead of on hashes.

The problem

  1. 01

    Explorer output is not context. An agent handed a raw transaction list has to infer what "0xa9059cbb" did on which protocol; it will guess, and it will be wrong often enough to matter.

  2. 02

    Building an indexer to feed an agent is the wrong project. Protocol adapters, price capture and contract labelling are a data-infrastructure company's core work, not a feature of your agent.

  3. 03

    Heavy wallets overflow the window. Hundreds of tokens and positions across chains arrive as thousands of lines; without truncation and scoping, the agent slows down or drops data.

What Datai delivers

  • MCP layer. An open-source MCP server with eight tools (all positions, positions by chain / chains / protocol, DeFi balances by chain, overall balance all-chains / by chain, token balances by chain) for Cursor, Claude Desktop, Eliza OS or any MCP client; a debug-truncation mode that preserves structure while capping size; chains exposed today: Ethereum, Arbitrum, Polygon, Avalanche, BSC, Base, Optimism.

  • User Insights via REST. The same positions and balances plus wallet analytics (beta): activity breakdown, protocols and chains used, DeFi, bridging and swap activity, net worth and transaction counts — the profile fields an agent needs to personalise advice or gate actions.

  • Transaction classification via REST. Labelled history (txClassification, txAction, protocol, functionName) for agents that explain, audit or summarise what a wallet did. REST only — not exposed as MCP tools today.

  • Format. Consistent JSON across chains and protocols; protocol commonName and token symbols included so responses can be narrated without a lookup.

Outcome

An agent that understands a wallet's DeFi, token and NFT holdings without you building an indexer — the outcome the MCP server is built for. The README's own example is a wallet copilot: user connects a wallet, the agent calls get_all_defi_positions, then summarises holdings, flags exposure and recommends actions. No accuracy or latency figures are claimed.

“Our experience with Datai’s API has been excellent, with a responsive team, expanded coverage, and reliable technology that seamlessly fits into our workflow.”

Evgeny YurtaevCEO, Zerion

Integration notes

  1. Start with get_all_defi_positions for a full scan and switch to chain- or protocol-scoped tools in production prompts; set DATAI_LIMIT low and enable DATAI_DEBUG_TRUNCATE while developing against large wallets.

  2. Keep the API key in the MCP server's environment, never in the model's context; one key serves both MCP and REST.

  3. For behavioural personalisation, provision the wallet once via REST (v2/userTx/provision), then read v2/userTx/analytics into the agent's memory rather than calling it per turn.

See it on your own wallets first.

Sign up by email, get 4,000 free credits, and run any endpoint against the addresses you care about.

Get API key (opens in a new tab) Contact sales

View the MCP server on GitHub (opens in a new tab)