Barbatos

Multiplatform Runtime Debugger & MCP Server.
"Understanding the language of systems, revealing hidden states, and reconciling the gap between developer and machine."

Using Barbatos through the MCP server to explore an app's structure and inspect objects:

The Terminal User Interface (TUI) demo:

Why the name "Barbatos?"

The name is inspired by the mythological Barbatos, but in the context of debugging.

Barbatos is a spirit (wikipedia) that grants the ability to understand the spoken language of animals. He reveals hidden treasures that have been concealed by the enchantment of magicians, gives knowledge of past and future events, and reconciles disputes between friends and those who hold power.

This project aims to grant the hability to understand deeply about an App, reveal hidden states in the heap, gives knowledge about changes of state (only past and present tho), and reconcile the gap between developers and machines (apps) by providing a more intuitive way to interact with them.

What is this project?

here you will find 2 things: A TUI debugger ("barbatos") and a core bridge ("barbatos-bridge") that includes both the Http API (for TUI) and the MCP server (for AI agents).

The focus of this project is to bring Frida powerfullness closer to devs and vibe-coders without the need to know how Frida is setup and how it works.

Why use Barbatos?

"Why use Barbatos if you can use Frida directly?"

Frida has a trade-off, in one end you got a very powerful tool, and in the other you have to understand how to setup (if you dont got a rooted device), and interact with its API. For some people this isnt a problem at all, but it does have a learning curve.

"And what about Objection?"

Ok thats a pretty good question, objection its an abstraction of frida just like barbatos, and it got a lot more capabilities, battle tested... For most cases objection is enough. But objection inherited the Frida overhead, he assumes you already has frida setup... which sometimes its enough to not use it at all. Also, objection is more focused on pentesting and red teaming.

unlike objection, barbatos was developed based on the assumption that the user know nothing about Frida to use, the only requirement is the app being debuggable, nothing more. Barbatos was designed with the mindset of being a debugging tool, so its more focused on that experience.

"Nowadays you can use a frida-mcp to handle the complexity"

Yes, all frida-mcp's (there are a few of them at this time) handles very smoothly actually, the only downside is that it also inherited the overhead of the user needing to know how to setup Frida, which again, can be a problem for some, but not for all.

Barbatos is a TUI debugger designed to be frictionless, 0 setup needed. It handle all the overhead needed to use Frida in your app. Apart from that, it has the same purpose of provide Frida's power in a more user-friendly way.

Main Use Cases

• Terminal-First Experience: ready to use in any terminal environment (local, SSH, WSL) without IDE plugins or complex setup.
• Simple App Exploration: quickly list classes, inspect objects, and understand app structure.
• Real-time App Debugging: modify field values on the fly and hook methods to see live execution flow.
• AI-Assisted Debugging: Use the built-in MCP Server to let LLMs (Claude/Gemini/Cursor) debug your app autonomously.
• Intercept Method Calls: Modify return values or arguments of any method.

Main Features

• Class Discovery: Real-time enumeration of loaded Java/Kotlin classes with package filtering.
• Deep Inspection: Recursive traversal of object hierarchies (Fields, Maps, Collections, Arrays).
• Zero-effort Integration: No IDE plugins or dependency installs — just run the binary and connect.
• Method Hooking: Intercept execution flow, inspect arguments, and capture return values.
• Live Field Editing: Modify primitive field values (String, Int, Boolean) in real-time to test hypotheses.
• MCP Server: Native Model Context Protocol support to connect your debugger to AI agents.

AI-Driven Debugging (MCP Server)

Barbatos provides native support for the Model Context Protocol (MCP). By connecting an AI agent to the Barbatos MCP Server, your LLM becomes an autonomous debugging assistant capable of direct runtime interaction.

Example Agent Capabilities (experimental):

• State Exploration: "Locate the active LoginState instance and tell me the value of its errorCode field."
• DI Troubleshooting: "Search for MyUseCase instance in MyViewModel and see if it is the same one in MyOtherViewModel."
• Method Hooking: "Tell me the parameters passed to the MyViewModel.fetchData method every time it is called."
• Live Edit: "Change the isPremium boolean on the active UserProfile instance to true."
• Root Cause Analysis: "List all instances of ApiError, check their message field, and identify recurring failure patterns."
• Mock Behavior: "Intercept the fetchData method in MyViewModel and modify its return to throw an exception when the isActive field is false."

Architecture

Barbatos uses a multi-stage pipeline for reliable communication:

graph TD
    A[Native TUI] -->|Ktor/HTTP| C[Python Bridge]
    B[AI Agent] -->|MCP Stdio| C
    C -->|Frida RPC| D[Frida JS Agent]
    D -->|Java Bridge| E[Multiplatform Runtime]
    E -->|ART| F[Target App]

1. Native TUI: Standalone Kotlin Native binary for a deterministic terminal experience.
2. AI Agent: Integration with AI agents (Claude, Cursor) via Model Context Protocol.
3. Python Bridge: Unified mediator exposing both JSON-RPC HTTP and MCP interfaces.
4. Frida Injection: JS agent injected into the process for runtime interaction.

Available MCP Tools

Installation

Quick Install (Recommended)

Run the following command to automatically detect architecture and setup environment:

curl -sSL https://barbatos.victorlpgazolli.dev/install.sh | bash

APT Repository (Debian/Ubuntu/WSL)

For Debian-based systems, you can use our APT repository to keep the tools up to date.

1. Add the GPG key:

   curl -fsSL https://barbatos.victorlpgazolli.dev/barbatos-repo.gpg \
     | sudo gpg --dearmor -o /etc/apt/keyrings/barbatos.gpg
   

2. Add the repository:

   echo "deb [signed-by=/etc/apt/keyrings/barbatos.gpg] https://barbatos.victorlpgazolli.dev/ stable main" \
     | sudo tee /etc/apt/sources.list.d/barbatos.list
   

3. Install the components:

   sudo apt update
   sudo apt install barbatos barbatos-bridge
   

MCP Server Setup

To use barbatos with Claude Desktop, add this to your claude_desktop_config.json:

{
  "mcpServers": {
    "barbatos-debugger": {
      "command": "barbatos-bridge",
      "args": ["mcp"]
    }
  }
}

Note: The unified bridge automatically handles Frida connection for the MCP server.