Search projects, hackathons, sponsors and tech

    ← All projects

    Tessera

    Tessera: An MCP marketplace where AI agents pay for API tools per-call using Hedera & x402.

    Demo

    About

    Tessera is a decentralized, per-use tool marketplace built specifically for autonomous AI agents using the Model Context Protocol (MCP). Today, the infrastructure of the web is stuck in human-centric patterns; when an AI agent needs to access a paid data source or run an API calculation, it cannot autonomously create a SaaS account, agree to a subscription, or enter a credit card. Tessera solves this by creating a true machine-to-machine economy. Instead of human signups, an agent connects to the Tessera MCP gateway, programmatically requests a tool, receives a per-invocation price quote, and instantly streams a micro-payment using the x402 protocol settled on Hedera. Once the payment is verified, Tessera securely executes the API call and feeds the clean data payload back to the agent in milliseconds. We chose Hedera because its fixed, ultra-low transaction fees make sub-cent micro-payments economically viable, unlocking a highly composable, zero-friction web where agents can seamlessly buy exactly what they need, one API call at a time.

    How it's made

    Tessera is built as an extensible Model Context Protocol (MCP) gateway using Node.js and TypeScript. At its core, we mapped the standard JSON-RPC tool-calling cycle of MCP directly onto the open x402 payment standard. When an AI agent invokes an MCP tool, the Tessera proxy intercepts the request, assesses the metered cost, and intercepts the flow to throw a standard cryptographic payload via the x402 scheme. We deeply integrated Hedera as our foundational settlement layer, leveraging the @x402/hedera reference packages alongside the open-source Blocky402 facilitator on the Hedera Testnet. Hedera provides immense benefit to this architecture; its fixed, predictable, and ultra-low $0.001 network transfer fees are what actually make sub-cent micro-payments economically viable for rapid-fire API inference. A particularly notable, "hacky" engineering choice was how we bypassed traditional web authentication. Because AI agents lack persistent browser sessions or traditional Web3 extension injections (like MetaMask), we built a lightweight transport middleware that wraps the x402 402 Payment Required header challenge into an inline MCP error response framework. This allows the client-side agent framework to catch the payment request, sign an off-chain transaction authorization autonomously via its private key, and pass it back to the facilitator, completing the entire transaction, validation, and API data delivery natively within a single, continuous command-line routine without breaking the LLM’s execution context