Orchestrating Multi-Agent Workflows with Buzz
Lukas Margeriego watch the original →
the gist
Buzz is an open-source collaboration platform that allows users to run multiple AI agents—such as Claude Code and Cursor—in a shared workspace to execute complex, multi-step design and engineering tasks.
Centralized Agent Orchestration
Buzz provides a Slack-like interface for managing AI agents, allowing developers to move away from fragmented workflows where context must be manually copied between separate IDEs and browser tabs. Users can create custom agents from scratch, selecting specific underlying models like Claude Code or Cursor, and assign them to shared channels or teams. This setup enables agents to communicate with each other, share project context, and execute tasks concurrently.
Multi-Agent Design Engineering Workflow
The author demonstrates a complex workflow by chaining specialized agents to redesign a blog page:
- Context Sharing: A 'Promptive Builder' agent connected to Cursor provides the codebase context to a 'Fizz' agent (running Claude Code).
- Component Integration: The Fizz agent pulls UI components from the BYQ component library via an MCP server to build the page structure.
- Visual Generation: A Higgsfield agent generates specific imagery based on descriptions provided by the Fizz agent.
- Canvas Rendering: The final design is pushed into MagicPath, an infinite canvas tool, allowing for real-time visualization of the agent-generated code.
MCP Configuration and Troubleshooting
To integrate new tools, users must add custom connectors by providing a remote MCP server URL. If an agent fails to recognize a newly added MCP, the author recommends stopping the agent in the settings menu and re-summoning it within the channel to force a refresh of its available tools and capabilities.