RAA
Provides Android application reverse-engineering and intelligence: ingests APK/AAB/APKS/XAPK archives, decompiles Dalvik/native code, extracts UI hierarchies and API contracts, and supports security auditing.
Provides a Burp Suite bridge for dynamic and API security testing and traffic inspection workflows.
Provides native reversing capabilities including C/C++ disassembly, JNI mapping, and dynamic symbol analysis.
Supports Dart AOT string scraping as part of Flutter application reverse engineering.
Supports Flutter app reversing, including BoringSSL TLS unpinning and Dart AOT string scraping.
Advertises HarmonyOS/ArkTS support as a multi-platform application intelligence target.
Advertises iOS/IPA support as a multi-platform application intelligence target.
Generates Jetpack Compose UI code and supports semantic Compose modernization from extracted view hierarchies.
Supports .NET/Xamarin assembly inspection for hybrid-framework reverse engineering.
Maps security audits to OWASP MASVS/MASTG and provides a MASVS checklist copilot.
Exports discovered API endpoints and synthesized specs as Postman Collection v2.1.
Generates SPDX SBOM reports for supply-chain and enterprise security compliance.
Provides dumping and inspection of on-device SQLite databases and app storage.
Supports Unity IL2CPP metadata dumping and inspection for Unity-based apps.
This server does not currently have a deployable release on Glama. Follow the repository's setup instructions to install and configure it for your MCP client, including any required credentials and local dependencies.
Overview
RAA (Reverse Android Anything) is a comprehensive, evidence-first Android reverse-engineering and application intelligence platform. It does not merely decompile APK files: it understands how an Android application is structured, how its Dalvik/native code executes, how its network APIs and databases interact, how its UI is composed, and how its original interface can be reconstructed and redesigned into modern frameworks.
LEGAL DISCLAIMER & LIMITATION OF LIABILITY ā PLEASE READ BEFORE USE
The developers, creators, maintainers, and contributors of RAA accept NO legal responsibility or liability of any kind for any damages, legal actions, criminal or civil liabilities, service disruptions, or copyright/intellectual property infringements resulting from the use or misuse of this software.
The end-user is exclusively liable and bears 100% legal responsibility for all actions performed with RAA. You must independently verify and ensure full compliance with all applicable local, national, and international laws, regulations, terms of service, and authorization requirements before inspecting, decompiling, analyzing, modifying, or intercepting any application or device.
RAA ingests APK, AAB (Android App Bundle), APKS, and XAPK archives, inventories their contents, detects frameworks and commercial packers, extracts view hierarchies and API contracts, and returns evidence-backed findings.
It is architected as one unified analysis core exposed simultaneously across:
Model Context Protocol (MCP): Dynamic stdio and Streamable HTTP servers for AI coding agents (Claude Code, Cursor, Antigravity, VS Code, Codex).
REST Surface: High-throughput FastAPI application with production security headers and OpenAPI 3.1 specifications.
Native CLI: Command-line tool suite built with Typer and Rich for CI/CD automation and terminal analyst workflows.
Python SDK: Direct programmatic library API.
Agent Skills: 11 versioned prompt-based workflows in
skills/.
Related MCP server: APK-MCP
Key Documentation
š Documentation Hub: Complete navigation guide and documentation index.
šļø System Architecture: Architectural principles, subsystem topology, runtime flow, and multi-agent coordination.
šØ System Design & Specification: Evidence-first truth model, fail-closed policy, 8pt UI design system, and API studio design.
š¤ Specialized Agents & Skills: Multi-agent orchestration, specialized subagent fleet (41 roles), and 11 discoverable skills.
š Strategic Roadmap: Upcoming feature highlights and planned capability milestones across future releases.
š¢ Enterprise Deployment Guide: Production hardening, multi-tenancy quotas, air-gapped installation, and disaster recovery.
š Security Architecture & Threat Model: Threat model, fail-closed scopes, tamper-evident audit logs, and OWASP MASTG compliance.
š¦ Release History & Changelog: Detailed version history and progression from Phase 1 to Phase 14 (v2.0.0 Production).
Core Capabilities
mindmap
root((RAA Engine))
Decompilation & Source
JADX Tier-1 Engine
Control-Flow Graphs
Method AST Recovery
Cross-Reference Tracing
Pure-Python Dalvik Disassembly
Hybrid Frameworks
Flutter TLS Unpinning
Dart AOT String Scraping
React Native Hermes Unpacking
Unity IL2CPP Metadata Dumping
.NET Xamarin Assembly Inspection
UI Extraction & Redesign
Runtime ADB Screen Capture
UIAutomator View Hierarchies
Chrome DevTools Protocol (CDP)
Jetpack Compose Code Gen
React Native Code Gen
Perceptual Visual Diffing
API Studio
Network Client Detection
URL Parameterization
OpenAPI 3.1 Spec Synthesis
Postman Collection v2.1 Export
Dynamic Instrumentation
Frida Script Injection
Frida Gadget ELF Patching
ADB Device Lease Manager
SQLite & Storage Dumping
Controlled Rebuild
Smali & Bytecode Patching
Network Security Bypass
Test APK Signing
Behavioral Verification
Enterprise Hardening
SHA-256 Hash-Chained Audit
Digital Signature Registry
CycloneDX & SPDX SBOM
Project Backup & Restore
Multi-Tenancy QuotasTool Registry (244 Implemented Tools)
RAA organizes its 244 tools into 7 distinct capability profiles:
Profile | Tool Count | Primary Use Case |
| 12 | Lightweight triage, manifest inspection, and basic artifact ingestion for small local LLMs. |
| 135 | Default static analysis, on-device PRoot mobile workspace harness, search-first MCP capability index, WASM binary decompiler, on-device AI reverser, Exploit PoC validator, MASVS checklist copilot, developer email OSINT footprint, class search, resource decoding. |
| 169 | Core tools + Headless VirtualDisplay runner, multimodal mobile GUI agent crawler, visual grounding touch navigation, Burp Suite MCP bridge, P2P encrypted device tunnel, passive traffic inspector, and Wireless ADB fleet discoverer. |
| 139 | Core tools + Cloud rotating gateway proxy, Burp Suite MCP bridge, passive traffic inspector with PII redaction, nomore403 bypass engine, CrowdSec CTI reputation check, SafeLine semantic WAF audit, and OpenAPI synthesis. |
| 140 | Core tools + Multimodal mobile GUI agent crawler, visual grounding touch navigation, Semantic Compose modernization, layout parity diff, Adaptive Icon Studio, banner generator, and AX-tree pruner. |
| 19 | Zero-trust remote MCP gateway configurator, Bug Bounty report & VRT formatter, engine verification, tamper-evident audit verification, SBOM generation, disaster recovery, and telemetry. |
| 244 | Complete, unrestricted superset of all registered capabilities. |
Specialized AI Agents & Skills
RAA coordinates 8 specialized AI agents via the Coverage Tracker Service (agent_coverage_get), supported by 11 discoverable skills in skills/:
skills/
āāā raa-apk-triage/ # Manifest, permissions, framework and entry activity triage
āāā raa-source-reconstruct/ # Decompilation, method call graphs, and AST recovery
āāā raa-ui-redesign/ # View hierarchies, runtime capture, and 8pt Compose redesign
āāā raa-api-discovery/ # Endpoint clustering, OpenAPI synthesis, and Postman export
āāā raa-flutter-reversing/ # Flutter BoringSSL TLS unpinning and Dart string scraping
āāā raa-native-reversing/ # JNI function mapping, dynamic symbols, and C/C++ disassembly
āāā raa-security-audit/ # MASVS compliance, secret scanning, and privacy data flows
āāā raa-controlled-rebuild/ # APK patching, gadget injection, signing, and ADB installation
āāā android-triage/ # Fast single-pass APK triage workflow
āāā platform-check/ # Host environment, dependencies, and sidecar diagnostics
āāā report-and-export/ # Executive Markdown, JSON, and PDF report generationInstallation & Configuration
RAA can be deployed in seconds across any environment ā from zero-install CLI execution to full IDE AI agent integration.
Quick Decision Matrix
flowchart TD
Start([š Getting Started]) --> Choice{How do you want to use RAA?}
Choice -->|Instant / No Setup| Zero[ā” uvx / pipx]
Choice -->|Active Development| Local[š» Python venv / pip -e .]
Choice -->|AI Coding Agent / IDE| MCP[š¤ MCP Server Integration]
Zero --> RunCli[Execute CLI commands instantly]
Local --> DevWork[Full codebase access & custom extensions]
MCP --> ClientType{Select Client}
ClientType -->|Claude Desktop / Code| ClaudeCfg[Configure claude_desktop_config.json]
ClientType -->|Cursor / VS Code| IdeCfg[Configure .cursor/mcp.json or settings.json]
ClientType -->|Google Antigravity| AgCfg[Configure mcp_config.json]
ClientType -->|Windsurf / Zed / Codex| GenericCfg[Configure IDE MCP settings]1. Installation Options
Execute RAA commands immediately without manually managing Python virtual environments:
# Using Astral uvx (sub-second execution):
uvx --from . raa --help
# Or via PyPA pipx:
pipx run --spec . raa --help# 1. Clone repository & enter workspace
git clone <repository-url>
cd RAA
# 2. Create and activate virtual environment
python -m venv .venv
# On Windows (PowerShell):
.\.venv\Scripts\Activate.ps1
# On Linux / macOS:
source .venv/bin/activate
# 3. Install in editable mode with development dependencies
pip install -e ".[dev]"
# 4. Verify installation
raa --version# Install as a permanent isolated system tool
uv tool install .
# Or via pipx:
pipx install .2. Workspace & Environment Configuration
Initialize the local analysis workspace and configure runtime parameters:
# Initialize local SQLite workspace (defaults to .raa-workspace/)
raa init
# Copy example environment configuration
cp .env.example .envVariable | Default | Purpose & Description |
|
| Directory for SQLite DB, evidence, cached DEX, and artifacts. |
|
| Active MCP profile ( |
|
| Logging verbosity ( |
|
| When |
|
| Bind address for REST API and Streamable HTTP MCP server. |
|
| Network port for HTTP/SSE services. |
| (None) | Bearer authentication token for securing REST and remote MCP access. |
|
| Maximum artifact upload/ingest size in bytes (default: 2 GB). |
3. AI Coding Agent & IDE MCP Integrations
RAA communicates with all major AI coding environments via the Model Context Protocol (Protocol 2026-07-28) using standard I/O (stdio) or Streamable HTTP (SSE).
graph LR
subgraph Clients["Supported MCP Clients & IDEs"]
CD["Claude Desktop"]
CC["Claude Code"]
CUR["Cursor"]
VSC["VS Code / Codex"]
AGY["Google Antigravity"]
WIN["Windsurf / Zed"]
end
subgraph RAA["RAA Core Engine"]
MCP_STDIO["stdio Transport (Local)"]
MCP_SSE["Streamable HTTP / SSE (Remote)"]
REG["244 Tools & 7 Profiles"]
end
CD --> MCP_STDIO
CC --> MCP_STDIO
CUR --> MCP_STDIO
VSC --> MCP_STDIO
AGY --> MCP_STDIO
WIN --> MCP_STDIO
MCP_STDIO --> REG
MCP_SSE --> REGAutomated Setup (One-Click)
Generate configuration automatically for your preferred AI client:
# Claude Desktop
raa setup --agent claude-desktop
# Cursor
raa setup --agent cursor
# Claude Code
raa setup --agent claude-code
# VS Code / Codex
raa setup --agent vscodeManual IDE Configuration Snippets
Path: %APPDATA%\Claude\claude_desktop_config.json (Windows) or ~/Library/Application Support/Claude/claude_desktop_config.json (macOS)
{
"mcpServers": {
"raa": {
"command": "python",
"args": ["-m", "raa.cli.main", "--workspace", ".raa-workspace", "--profile", "full", "serve", "mcp"]
}
}
}Path: Project root .cursor/mcp.json or Cursor Global Settings:
{
"mcpServers": {
"raa": {
"command": "python",
"args": ["-m", "raa.cli.main", "--workspace", ".raa-workspace", "--profile", "core", "serve", "mcp"]
}
}
}Path: Global %USERPROFILE%\.gemini\config\mcp_config.json or Workspace .agents/mcp_config.json:
{
"mcpServers": {
"raa": {
"command": "python",
"args": ["-m", "raa.cli.main", "--workspace", ".raa-workspace", "--profile", "full", "serve", "mcp"],
"env": {
"RAA_LOG_LEVEL": "INFO"
}
}
}
}Path: .vscode/settings.json or User Settings:
{
"mcp.servers": {
"raa": {
"command": "python",
"args": ["-m", "raa.cli.main", "--workspace", ".raa-workspace", "--profile", "core", "serve", "mcp"]
}
}
}Path: ~/.codeium/windsurf/mcp_config.json or Zed settings.json:
{
"mcpServers": {
"raa": {
"command": "python",
"args": ["-m", "raa.cli.main", "--profile", "core", "serve", "mcp"]
}
}
}For shared servers, remote containers, or web-based AI clients:
# Start RAA MCP in Streamable HTTP / SSE mode:
raa serve mcp --transport sse --host 0.0.0.0 --port 8765Connect using standard SSE URL: http://<server-ip>:8765/sse
4. š³ When is Docker Required?
Docker is 100% OPTIONAL for typical RAA workflows. All 244 core static analysis, decompilation, UI reconstruction, and API synthesis tools run in pure Python without Docker.
graph TD
Op[Analysis Request] --> Type{Task Category}
Type -->|Static Analysis / DEX / AST| NoDock[ā
Pure Python Native ā No Docker Required]
Type -->|UI Reconstruction / Compose| NoDock
Type -->|API Synthesis / Network AST| NoDock
Type -->|Physical Device ADB / Frida| NoDock
Type -->|Headless Cloud Android Emulator| Dock[š³ Docker Required: redroid / budtmo]
Type -->|Untrusted Malware Sandboxing| Dock[š³ Docker Required: Ephemeral Container]
Type -->|Multi-Tenant Cloud CI/CD Engine| Dock[š³ Docker Required: Isolated Container Pool]Docker is strictly required only in the following 3 conditions:
Scenario | Condition | Docker Tool / Image |
1 | Headless Virtual Android Emulators | When running dynamic tests without physical Android hardware connected ( |
2 | Untrusted Malware Sandboxing | When executing untrusted binary payloads or aggressive runtime hooking in an isolated sandbox ( |
3 | Cloud Multi-Tenant CI/CD Fleets | When running RAA inside enterprise Kubernetes or containerized microservices to guarantee zero-residue workspace isolation. |
5. Starting Services & Health Verification
# 1. Verify environment, sidecars, and dependencies
raa doctor
# 2. Start REST API (Swagger UI at http://127.0.0.1:8765/docs)
raa serve api --host 127.0.0.1 --port 8765
# 3. Start interactive MCP server (stdio)
raa serve mcpQuick Start: CLI Workflows
# Ingest an APK and create a project
raa project create "My Analysis Project"
raa artifact ingest prj_xxx /path/to/target.apk
# Inspect Manifest & Decompile
raa manifest inspect prj_xxx art_xxx
raa source search prj_xxx "AuthToken"
# Synthesize OpenAPI 3.1 Specification and Export Postman Collection
raa api synthesize prj_xxx art_xxx
raa api postman prj_xxx art_xxx --output ./postman_collection.json
# Extract UI, Capture Runtime Screen & Generate Jetpack Compose Redesign
raa ui tree prj_xxx art_xxx
raa ui redesign prj_xxx screen_main --framework compose
# Verify Tamper-Evident Audit Trail & Generate CycloneDX SBOM
raa admin audit-verify
raa admin sbom --format cyclonedx
# Backup Project for Disaster Recovery
raa admin backup prj_xxx --output ./backup.raa.bundle.tar.gzFrequently Asked Questions (Q&A)
The end-user bears 100% sole legal liability. The developers, authors, maintainers, and contributors of RAA accept no responsibility, obligation, or liability of any kind for any legal consequences, damages, civil suits, regulatory penalties, or criminal charges arising from your use or misuse of this tool. You must ensure you have explicit written authorization before analyzing, decompiling, or intercepting any target software or hardware.
RAA is a universal application reversing platform supporting:
Android: APK, Android App Bundle (AAB), split APK sets (APKS), and XAPK archives across ARM64, ARMv7, x86, and x86_64.
Apple iOS: Application containers (
.ipa), 64-bit Mach-O binaries, FairPlay DRM audit (cryptid), Info.plist permissions, and PKCS#7 signed embedded provisioning profiles.Huawei HarmonyOS & OpenHarmony:
.happackages, Stage (module.json) and FA (config.json) application models, and ArkTS Panda ABC bytecode (classes.abc).
RAA natively implements the Model Context Protocol (MCP) specification. You can run RAA as a local MCP server over standard input/output (stdio) or as a networked background server over Streamable HTTP (FastMCP). AI agents discover all 244 registered tools with their complete JSON Schema inputs and execute end-to-end reversing, security audits, and UI reconstruction autonomously.
Yes, absolutely. RAA is engineered with a strict local-first, zero-egress architecture. All core analysis routines, Dalvik/DEX parsing, Mach-O inspection, and ABC bytecode disassembly are written in pure Python with zero external network dependencies. No telemetry or data is ever transmitted outside your local machine.
RAA enforces a fail-closed security policy engine. Potentially hazardous operations (such as binary patching, code injection, or device mutation) require explicit project-level scope grants (e.g., android.patch, dynamic.hook, dynamic.device). Any operation lacking the authorized scopes fails closed with a PolicyDeniedError before execution. In addition, all tool invocations are permanently cryptographically recorded in an append-only, hash-chained audit log.
RAA uses capability profiles to tailor tool exposure to specific agents and contexts:
minimal (12 tools): Essential triage and summary tools for lightweight models.
core (135 tools): Static analysis, source inspection, frameworks, and decompilation.
dynamic (169 tools): Runtime ADB, Frida instrumentation, and network interception.
api (139 tools): API reverse-engineering, spec synthesis, and threat intelligence.
ui (140 tools): View hierarchy extraction, visual grounding, and UI code generation.
admin (19 tools): Cryptographic audit verification, doctor, SBOM, and licensing.
full (244 tools): Complete, unrestricted tool catalog for frontier models.
No, root is not mandatory. While rooted devices unlock full kernel tracing and raw socket capture, RAA supports non-rooted workflows using:
Privileged operations via the Shizuku IPC Bridge (shizuku_privileged_exec).
Standalone userspace Linux containers via Termux PRoot (termux_proot_workspace_manage).
Virtual headless fuzzing via VirtualDisplay and Docker-Android emulators.
RAA extracts runtime view hierarchies via ADB or Chrome DevTools Protocol (CDP), prunes the accessibility tree (AX-tree), clusters styling properties (typography, paddings, colors) against an 8pt design system, and synthesizes production-ready Jetpack Compose (Android) or React Native code with perceptual visual diff verification.
Verification & Quality Gates
RAA maintains 100% test pass rates and strict zero-lint compliance:
# Run the complete test suite (437 tests)
python -m pytest -v
# Run linting and style checks
python -m ruff check src tests
python -m ruff format --check src testsLicense & Legal Disclaimer
Limitation of Liability & User Responsibility
IMPORTANT LEGAL NOTICE: USER ASSUMES ALL LIABILITY
No Developer Responsibility: The authors, maintainers, and contributors of RAA (Reverse Android Anything) provide this software "AS IS" without warranty of any kind. Under no circumstances shall the developers be held liable for any direct, indirect, incidental, special, exemplary, or consequential damages, or any legal liabilities, civil claims, regulatory penalties, or criminal prosecutions arising from the use or misuse of this software.
Sole End-User Liability: The user bears 100% sole legal liability and responsibility for all reverse-engineering, dynamic instrumentation, traffic analysis, decompilation, and testing activities conducted using RAA. Users must independently ensure that all actions comply strictly with all applicable local, national, and international laws, regulations, intellectual property protections, contracts, and software licenses (such as the DMCA, CFAA, and EU Computer Misuse directives).
Authorized Use Only: RAA is intended exclusively for authorized security audits, academic research, and defensive evaluation of applications owned by the user or where explicit, written authorization has been granted by the rightful owner.
For full security and compliance policies, refer to SECURITY.md.
License
This project is licensed under the MIT License. See the LICENSE file for complete details.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelogā¦
This server cannot be deployed
Maintenance
Related MCP Connectors
Turn websites into Android apps from your AI agent: create, set icons, build and download APK/AAB.
151Give your AI agent a real cloud Android phone: push APKs, drive it, get screenshots and logs.
Cloud Android phones for AI agents: create a phone, read the screen, tap, type, install APKs, park.
Code intelligence platform for AI agents. 20 tools for architecture, security & impact analysis.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceEnables analysis of native Android libraries (SO files) and Flutter applications, including decompilation, function identification, SSL certificate bypass patching, and binary modification through 30+ specialized tools.12-
- AlicenseNot gradedqualityBmaintenanceProvides AI agents with a unified MCP interface for Android APK reverse engineering, metadata analysis, Smali/resource editing, rebuilding, signing, verification, and ADB device testing.MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to inspect and decompile software they have no source for ā Android APKs, Electron bundles, .NET assemblies, and Go, Rust or WASM binaries ā plus adb device interaction, through a token-guarded Streamable HTTP MCP endpoint, stdio MCP, or a CLI. Ghidra-powered native binary decompilation can be opted in at startup for ELF, PE and Mach-O targets.7Do What The F*ck You Want To Public
- AlicenseNot gradedqualityNot gradedmaintenanceEnables AI agents to inspect and reverse-engineer native binaries, JavaScript/Electron apps, Go and Rust binaries, WASM, EVM bytecode, SQLite databases, and recorded crashes through a single MCP interface that bridges tools such as Ghidra, Frida, Hopper, binwalk, and jadx. Returns evidence-backed findingsāpseudocode, symbols, strings, imports, routes, IPC boundariesāso agents can explain behavior and build similar features.MIT