Max, An OpenClaw Agent is Live on MultiversX

Meet Max, Our OpenClaw Agent Powered By The Universal Agentic Commerce Stack

For the past few weeks, we've been shipping the infrastructure for agentic commerce on MultiversX. Protocol integrations, identity standards, developer libraries. The full stack.

Today, we're shipping something you can try for yourself.

Our first demo of an autonomous agent is live on MultiversX devnet. It receives funds, makes its own trading decisions, executes swaps onchain, and sends the purchased tokens back to you. No human in the loop. End to end agentic interaction.

What the Agent Does

MultiversX Universal Agentic Commerce Stack

The final point of the demonstration is called Mystery Swap.

You connect your wallet and start a session with the agent. You choose how much to swap. The agent pulls live token data from xExchange, decides how to allocate your funds across multiple tokens, executes the swaps autonomously, and sends every purchased token back to your wallet.

You can also just chat with it. The agent runs a full conversational session. Ask it questions, request specific actions, or let it reason through the allocation on its own.

It’s a simple interaction. What's underneath shows a glimpse of what matters, and what’s unfolding around the world.

The Closed Loop Is the Benchmark

Most agents in crypto show one capability. An agent that reads onchain data. An agent that signs a transaction. An agent that responds to a prompt. Useful proof points, but not what agents need to do in production.

Production agents complete full economic cycles. Receive a task. Make decisions. Execute multiple onchain actions. Return results. Handle the money.

This is the first step that showcases the full end to end flow. The agent holds funds, reasons about allocation, executes a sequence of swaps against live liquidity, and settles assets back to the user. Every step is onchain and verifiable.

The question that matters isn't "can an agent do one thing." It's "can an agent do a job, executing more generalized tasks to achieve what matters." This one, shows the first signs it can.

What's Under the Hood - MultiversX Universal Agentic Commerce Stack

This demonstration runs on the Universal Agentic Commerce Stack we've been building on MultiversX. Each component handles a specific part of what the agent needs to do:

• ‍Discovery and state access through the Model Context Protocol (MCP) integration. The agent interacts with onchain data and execution functions through structured, typed interfaces. No scraping, no unstructured parsing, no bespoke integrations per model.‍
• Programmatic execution through the Agent Commerce Protocol (ACP), which standardizes how agents construct and complete transactions.‍
• Gasless operation through Relayed v3. The agent executes transactions without the user approving each one or managing gas. This is what makes multi-step agent workflows economically viable.‍
• Identity and accountability through MX-8004, MultiversX's implementation of the trustless agent standard. The agent has a soulbound onchain identity. Jobs are validated. Reputation is derived from verified work, not self-reported metadata.‍
• Settlement through x402, providing HTTP-native payment primitives for machine-to-machine transactions.

What This Means If You Use Agents

The Mystery Swap is simple by design. The primitives underneath it are not.

An agent that can receive funds, make decisions, execute onchain, and return results is an agent that can rebalance a portfolio. Claim and restake rewards. Execute a DCA strategy while you're offline. Manage positions across multiple protocols without you being online.

The pattern is always the same: you delegate a bounded task, the agent executes autonomously, and you get verifiable results onchain.

This is what agent infrastructure is actually for. Not chatbots that read your balance. Autonomous systems that do real economic work on your behalf.

What This Means If You Build Agents

The libraries used by Max, the agent, are all open source and available today.

If you can define a workflow as "receive input, make decisions, execute onchain, return results," you can build an agent application on MultiversX.

The OpenClaw template provides the starting point. The agentic commerce stack provides the rails for verifiable agentic interaction. The identity layer provides trust and accountability.

The Universal Agentic Commerce Stack in Action

This is a live demonstration with applicability to something much larger.

Over the past weeks, we have been shipping a complete protocol stack for agentic commerce. Discovery through UCP. Authorization through AP2. Execution through ACP. Settlement through x402. Identity and reputation through MX-8004. Structured state access through MCP.

The Mystery Swap is the first live demonstration of these components working together end to end. It won't be the last.

The standards that will define how agents operate economically are being written right now. UCP by Google. MCP by the Linux Foundation's Agentic AI initiative. AP2 by Google and major payment networks. ACP backed by OpenAI and Stripe. All of them are active specifications with real institutional backing, and they're converging fast.

MultiversX is building at that edge. Implementing these standards at the protocol layer, shipping working integrations, and putting them in front of users and builders around the world, while the rest of the industry is still publishing roadmaps.

The infrastructure for agentic commerce is not part of some distant hypothetical future project we intend to build.

‍It's here, now. It's open source. And it's running onchain. TODAY.

MEET MAX!

‍

All the code is live and open source.

Repositories:

• OpenClaw template: github.com/sasurobert/mx-openclaw-template-solution
• OpenClaw relayer: github.com/sasurobert/multiversx-openclaw-relayer
• OpenClaw skills: github.com/sasurobert/multiversx-openclaw-skills
• Starter kit: github.com/sasurobert/moltbot-starter-kit
• MX-8004 agent registry: github.com/sasurobert/mx-8004
• x402 integration: github.com/sasurobert/x402_integration 
• MCP server: github.com/sasurobert/multiversx-mcp-server 
• ACP adapter: github.com/sasurobert/multiversx-acp-adapter 
• Agent Explorer: agents.multiversx.com