ConduitX
Decentralized AI agent data vending machine with x402 metering & Hedera HCS receipts.
Demo
About
ConduitX is a decentralized data vending machine for autonomous AI agents. While agents have payment rails, they lack an open marketplace to discover, buy, and audit blockchain data query by query. With ConduitX, an agent discovers data providers, receives an HTTP 402 Payment Required challenge with dynamic pricing based on query complexity, and settles micro-payments in under two seconds over Hedera. The seller delivers live data from The Graph subgraphs and immediately publishes a cryptographic receipt to Hedera Consensus Service (HCS Topic 0.0.9840084). Future buyer agents read these public HCS receipts via Mirror Nodes to calculate provider reputation objectively before purchasing. ConduitX also includes a native Model Context Protocol (MCP) server so AI models like Claude and Cursor can procure blockchain data autonomously.
How it's made
ConduitX is built as a TypeScript monorepo combining Hedera Consensus Service, the x402 payment protocol, The Graph Subgraph Studio, and a Next.js 16 workstation dashboard. On the payment rail, we built custom Express middleware that intercepts incoming data queries, meters cost dynamically based on request complexity (base fee + per-row returned), and issues HTTP 402 challenges. The buyer agent settles micro-payments using the Hedera SDK via the Blocky402 facilitator with automatic ECDSA private key parsing. Once settled, the data provider executes live GraphQL queries against The Graph Subgraph Studio (Uniswap v3 DEX analytics, token concentration risk, and portfolio indexers) and immediately publishes an immutable receipt message to Hedera Consensus Service (HCS Topic 0.0.9840084). A notable design innovation is turning HCS receipts into a decentralized reputation oracle: instead of relying on a centralized database to rate providers, our broker queries Hedera Mirror Node REST APIs to deterministically compute provider reliability, latency, and failover routing directly from public consensus timestamps. We also implemented a native Model Context Protocol (MCP) server so Claude Desktop and Cursor can run this entire data procurement loop as a standard agent tool.