# FLUJO [Health: Active]

**Category:** 🤖 Coding Agents  
**Repository:** https://github.com/mario-andreschak/FLUJO  
**GitHub Stars:** 665  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/flujo

## Description
Multi-Agent + Automation Harness: Graph-based Workflows, MCP, Self-Improving Agents. NextJs+React

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "flujo": {
    "command": "npx",
    "args": ["-y","flujo-ai@latest"]
  }
}
```

## Documentation

## What the FLUJO MCP server does

FLUJO combines model configuration, MCP connections, agent construction, and run inspection in a local workspace. It stores workspace data on the user's computer while requests sent to cloud models and connected services leave the machine according to the configured connections.

The FLUJO MCP server functionality centers on Connected Apps. Developers can attach MCP servers that expose files, browsers, APIs, and other services. Connections can be selected for individual agent steps, tested before use, and reused in chat, background automations, or external applications through FLUJO's local MCP proxy.

Agents can be created as sequential plain-language recipes or as visual graphs. Expert flows support AI, connected-app, static, subflow, start, and finish nodes. Graphs can branch, loop back to earlier steps, hand work to specialists, and call reusable subflows.

## How it works

First configure one or more model providers or local models, then add connected applications. MCP connections may come from curated apps, the MCP Registry, a remote URL, a GitHub repository, or a local command. A connection can expose tools, resources, and reusable prompts; supported connections may also provide roots or model sampling.

The tool inspector lets you view a connected server's declared capabilities, call tools, and inspect resources from its detail view. Saving a connection and successfully completing an operation are separate checks, so a configured server should be tested before it is attached to an agent.

Each agent step can receive its own model, instructions, tools, and resources. Runs can be examined through FLUJO's debugging interface, and compatible MCP Apps can display an interface inside chat when the connected server declares the required capabilities.

## Setup and configuration

The documented quick-install options include a Windows installer, a PowerShell command, and a shell command for Linux or macOS. Users with a supported Node.js 22 or 24 release can run the published package with `npx flujo-ai@latest`. Docker and building from source are also listed as installation choices.

After installation, open `http://localhost:4200` and keep the FLUJO process running. The installers download dependencies and require internet access. Some connected tools require their own runtimes or credentials. Model providers and saved connections must be configured separately, and provider access or usage limits still apply.

## Tools and capabilities

The FLUJO MCP server supports:

- Connecting MCP servers through registries, URLs, GitHub repositories, or local commands.
- Browsing and calling connected tools from a server detail view.
- Inspecting MCP resources and reusable prompts.
- Configuring roots and opting into model sampling where supported.
- Attaching MCP capabilities to specific agent steps.
- Building workflows with branching, loops, handoffs, and reusable subflows.
- Exposing a configured connection through FLUJO's local MCP proxy for another compatible client.
- Running agents in chat, automations, or external applications.

## Limitations and notes

FLUJO is local-first, but it is not isolated from network services by default. Cloud models and connected applications receive the requests sent to them. The behavior of interactive MCP Apps, roots, sampling, and other optional capabilities depends on what each connected server declares and supports.

The README does not identify specific external MCP clients by name. It refers generally to another compatible client using FLUJO's local proxy. Connected tools may need additional runtimes, credentials, or successful setup beyond saving their connection. Subscription and provider allowances continue to govern model access.

_Full upstream README: https://allmcps.com/mcp/flujo/readme_