Agent ecosystems need more than models - they need infrastructure. So we started Arcade Labs, a home for the experimental agent-infrastructure projects we want to build in the open. The first one out the door is Omni, and it tackles a problem I run into constantly: when an agent has access to thousands of tools, how does it find and use the right one?

The Tool Discovery Problem

This is the natural next chapter after the tool abstraction problem. Once you’ve built thousands of good tools, you can’t just hand all of them to a model - the schema balloons, context gets eaten, and selection accuracy drops. The agent needs a way to discover the right tool for a task, then execute it securely.

What Omni Is

Omni is an MCP server preloaded with 500+ productivity and work tools - Google Workspace, Microsoft 365, Zendesk, X/Twitter, and more. Instead of pre-loading every tool into the context window, an agent asks Omni for the capability it needs, Omni returns the right tool, and the agent executes the action across connected services - all over Arcade’s authorized, per-user tool calling.

In other words: find the tool, then do the work. I’ve been using it as the tool layer behind my own agents, and it’s the same building block I reach for when I want an agent to actually act in the outside world.

Try It

Arcade Labs is where we’ll keep shipping this kind of thing in the open. Omni is just the first.

For more about Arcade check out the website at https://www.arcade.dev/ and the GitHub https://github.com/arcadeai/