Three weeks ago, we kicked off a bold goal: build an open-source platform where your AI agents connect, collaborate and share context across systems, without rebuilding integrations each time. Built to be secure, scalable, and ready for enterprise use.
Behind the scenes, we have been prototyping this infrastructure using two key technologies: kagent and mcp-use.
kagent is an open-source orchestration layer that helps us deploy, scale and manage AI agents and tool servers, the engine room of your AI setup. You define your tools and agents in one simple configuration file. It automatically discovers and registers new tools, and includes a cockpit dashboard to monitor what is running and manage issues.
We have paired it with mcp-use, a lightweight protocol layer that helps AI agents interact with external tools in a consistent and scalable way. Agents can dynamically discover what tools are available and what inputs and outputs they expect. It supports real-time streaming via Server-Sent Events, and pluggable auth modules (API-key, OAuth2 client-credentials, JWT) so clients can talk to protected MCP endpoints out of the box.
By combining the two, we created a setup where agents can dynamically discover, invoke and scale tool servers, without the spaghetti of hardcoded integrations.
This is the groundwork for the cockpit we are building: modular, secure and open-source. A foundation companies can use to become truly AI-first, with less friction and more results.