Limit orders via MCP

As part of the hackathon, we are developing an open-source MCP module for limit orders, which can be integrated by external teams into their own AI agents. After the hackathon, we will incorporate this module in our own AI-agent called HAIA. Previously, we’ve already connected basic 1inch swap functionality to our assistant and plan to integrate Fusion+ swaps in the near future.

Our MCP server dramatically simplifies agent interaction with the 1Inch protocol across any blockchain network, making complex multi-chain DeFi operations as simple as natural language commands regardless of whether you're working with existing chains or the new networks 1Inch plans to integrate during the hackathon like Monad, Starknet, Cardano, and others.

Also what's really cool is that this universal connector brings 1Inch directly into ChatGPT, Claude Desktop, Claude Code, Cursor, and any other MCP-compatible tools, enabling seamless protocol access across the entire AI ecosystem without custom integration work.

We are transforming HAIA from a financial AI assistant into a dynamic platform capable of connecting and scaling any Web3 tools, agents, and protocols through a modular architecture. Our goal is to make Web3 agents as accessible as App Store apps, while maintaining full control over private keys, data, interfaces, and execution logic.

At the core lies an MCP-based protocol, enriched with data schemas, UI components, and interaction rules. Each agent or tool can be integrated as a standalone module, added dynamically without changing HAIA’s core code. This enables rapid expansion of functionality, the addition of new use cases, custom integrations, and white-label solutions for partners or enterprise clients.

This architecture turns HAIA into a “browser for agents” — not for websites like most current Web3 interfaces, but for specialized AI/Web3 agents. It allows teams and protocols to build and whitelist their own agents, making them accessible within HAIA’s secure environment — with native support for modular accounts, AI interfaces, and cross-chain infrastructure.

Modular Python backend, designed for integrating intelligent AI/Web3 agents via the MCP protocol.

On the frontend, we use React as the core framework, with Vite for lightning-fast bundling and development. Chakra UI powers the component library for clean, accessible design, while Zustand manages state with minimal overhead. For data fetching, we use both Axios and native Fetch, including support for streaming responses (SSE) to enable real-time AI interactions. Wallet integration is handled via thirdweb-react, giving us multi-chain support out of the box.

On the backend, the system is built in Python using FastAPI and FastMCP for a lightweight and performant web server. The architecture revolves around MCP (Model Context Protocol), allowing us to describe and serve dynamic tools/agents. Pydantic defines strict data schemas, while PyJWT handles authentication. We use Requests to interact with external APIs (like 1inch), and Asyncio ensures concurrency and speed.

For data storage, we rely on PostgreSQL as our primary database and Redis for caching and temporary state, especially important in conversational flows. During development, we also leverage LangGraph to test agentic workflows in an isolated graph-based environment.