The local-first agent builder with a debugger attached

What it does FLUJO is an open-source workspace for building multi-agent workflows and automations. It connects to various LLM providers and MCP servers, letting you assemble flows through either a simple recipe builder or a drag-and-drop node editor with branching, loops, and reusable subflows. You can chat with the resulting agents, automate them via triggers, or call them through an OpenAI-compatible API.

The interesting bit The framework is MCP-native rather than treating tools as an afterthought: it can install, inspect, test, and even proxy MCP servers to other clients like Claude Desktop or Cursor. It also ships with a visual debugger that lets you set breakpoints and inspect per-node state before and after each step.

Key highlights

• Visual flow builder supports branching handoffs, loops, and subflow nodes with isolated state
• Native MCP client and proxy: browse, test, and reuse configured servers in other apps
• Multi-model support including local Ollama, Claude Subscription plans, and OpenRouter
• Automation triggers: cron schedules, webhooks, file watchers, URL monitors, and MCP tool polling
• Human-in-the-loop tool approval and a built-in visual debugger for any provider

Caveats

• The MCP proxy endpoint is currently restricted to localhost
• No Ollama-compatible server endpoint for external clients; use the OpenAI-compatible API instead

Verdict Try it if you want a private, visual alternative to cloud agent platforms. Skip it if you are looking for a fully managed service.