AI agents are now powerful enough to navigate the internet on people’s behalf. They can make restaurant reservations, buy concert tickets, post on social media, and complete other tasks faster and more efficiently than humans ever could. Used properly, they can help individuals get more from their online experience.
World helps ensure that the agents navigating the internet are acting on behalf of a real, individual human, while helping to prevent bad bots from gaming online systems. That trust layer is becoming essential as a growing contingent of businesses adopt agentic payment protocols like x402, opening their services to autonomous agents that can now transact on a person's behalf.
AgentKit is a toolkit for building human-backed agents. Using World ID, it helps ensure humans get the upside of agents without the bad-bot downsides. Individuals give agents the power to act on their behalf with specific guardrails. And merchants and website operators can enable discounts or limited drops without worrying that it will all end up in the arms of bad bots.
AgentKit was designed for developers, but you don’t need to be one to use it. ToolRouter is a simple interface built on top of AgentKit that lets anyone connect a verified World ID to an agent. This guide walks through how to set it up. (If you’d rather register your agent with AgentKit directly, please read the developer docs.)
Before you start: What you’ll need
- A verified World ID (find an Orb here)
- World App on your phone
- An agent (AgentKit supports MCP clients like Claude Code, Codex, Cursor, VS Code, Hermes, and OpenClaw)
Step 1: Go to ToolRouter.
ToolRouter lets you manage your balance and agents in one simple familiar interface. Once you’re on the platform, you can delegate your World ID by verifying with World App via the prompts on the website.
Step 2: Get your API key.
Set up your API key. Your agent will need it to access websites that check for proof of human.
Step 3: Connect your agent.
Set up the MCP connection with your agent. ToolRouter gives you the copyable commands for tools like Claude Code, OpenClaw, Hermes, and more. Just insert your API key, copy the code, and paste it into your agent tool.
Step 4: You’re done. Now use it and it will automatically improve over time
Your agent now has a delegated World ID, meaning that any website or API that supports AgentKit will know it’s verified and acting on your behalf. As more services support AgentKit, ToolRouter will automatically add them giving your agent more super powers without requiring additional configurations.
AgentKit in action: the Lift Off hat drop
Agents are changing how we use the internet, and World ID helps ensure they don’t compromise the integrity of e-commerce.
At Lift Off, the team demonstrated this concept with a limited-edition merchandise drop. Five hundred "Human in the Loop" hats were made available exclusively to verified World ID holders.
The demonstration showed how a World ID-backed agent could discover the drop, verify eligibility, and complete the claim process on behalf of its human. Because the agent was delegated by a verified World ID holder, the storefront could recognize that it was acting on behalf of a real person rather than an anonymous bot.
To complete the task, the agent used Exa to search for the product and locate the storefront. It then navigated a Shopify storefront, applied the appropriate discount, and completed the order workflow. Tasks that would normally require a human to search for a product, switch between tabs, copy codes, and fill out forms were handled by the agent.
All 500 hats were claimed by verified individuals across multiple countries, including the United States, Germany, Japan, and the United Kingdom. Rather than being scooped up by a small number of actors running large bot networks, the drop was distributed among unique humans around the world.
現在、100ドル（16,000円相当）のWorld公式グッズ「Human in the loop」ハットを無料で入手できるキャンペーンが実施中。
— World日本公式（World Japan） (@worldnetworkjp) May 13, 2026
対象は、World IDのAgentKit（エージェントキット）で認証済みの、人間が利用するAIエージェントに限ります。
ぜひWorld IDの連携から購入体験までトライしてみてください。… pic.twitter.com/eiHPUsAEEt
Just as importantly, the same safeguards that enabled the purchase also prevented abuse. The hat drop was limited to one item per verified human. An agent could help a person claim their allocation, but it could not claim additional hats beyond that limit. Creating more agents didn't create more eligibility.
The demonstration also highlighted how businesses can introduce additional human-in-the-loop checks for sensitive actions. For example, a service could require a fresh World ID verification before allowing an agent to publish content, make a high-value purchase, or perform another action on a user's behalf.
The result is a model where agents can do more work for people while remaining accountable to the humans behind them.
What’s next for AgentKit
World is committed to ensuring everyone can benefit from the next wave of AI and agents, while keeping the internet protected from abuse. Keep an eye on AgentKit for additional services coming soon. If you register with ToolRouter, you’ll automatically get access.
