mario-andreschak/FLUJO

FLUJO is an ambitious and well-developed visual platform for building AI workflows. It brings models, MCP servers, tools, branching, loops, reusable subflows, automation triggers, and debugging into one interface. The visual debugger is a standout feature because users can pause a workflow, move through it one step at a time, and inspect what changed. Support for several model providers, Ollama, Claude subscriptions, OpenAI-compatible endpoints, and the official MCP Registry also makes the project flexible. The repository shows thoughtful security work. Secrets stay server-side and are encrypted locally, tool calls can require human approval, MCP Apps run inside restricted sandboxes, and the Docker setup binds to localhost by default. The documentation clearly warns users that FLUJO has no built-in authentication and should not be exposed publicly without a secure reverse proxy.

The biggest weakness is that the README has become too large. It explains nearly every feature, installation method, security detail, and advanced option on one page. That depth is useful, but it makes the first-time experience harder to follow. A short quick-start guide with one sample workflow would help new users reach their first successful run faster, while the detailed material could move into separate documentation pages. Contributor support also needs more structure. The README provides basic pull request steps, but the repository would benefit from dedicated contribution and security policies, architecture guidance, coding standards, and clear instructions for running linting, type checks, and tests. The one-line installation commands are convenient, though security-conscious users may prefer instructions for downloading and reviewing the scripts before execution. Overall, FLUJO is a capable, technically serious project with an unusually complete MCP feature set. Better onboarding and contributor documentation would make it easier for the project to grow beyond its current audience.