mobsf-mcp
Enables AI agents to perform autonomous Android security analysis, including static analysis, dynamic analysis, and Frida instrumentation, powered by MobSF.
README
mobsf-mcp
<p align="center"> <img src="assets/logo.png" alt="mobsf-mcp" width="400"> </p>
An MCP (Model Context Protocol) server that gives AI agents the ability to perform autonomous Android security analysis - static analysis, dynamic analysis, and Frida instrumentation - powered by MobSF.
Point your AI agent at an APK and let it autonomously discover vulnerabilities, inspect decompiled source, test exported activities, bypass SSL pinning with Frida, and generate security reports.
Architecture
┌─────────────────────┐ ┌──────────────┐ ┌─────────────┐
│ AI Agent │ MCP │ mobsf-mcp │ REST │ MobSF │
│ (Claude/Ollama/etc) │◄───────►│ Server │◄───────►│ (Docker) │
└─────────────────────┘ stdio └──────────────┘ HTTP └──────┬──────┘
│ ADB
┌──────▼──────┐
│ Android │
│ Emulator │
│ (optional) │
└─────────────┘
Static analysis (no emulator needed): Upload APK → decompile → analyze manifest, permissions, code, secrets, network config, crypto, certificates, trackers.
Dynamic analysis (emulator required): Install app on emulator → monitor runtime behavior → intercept API calls with Frida → test TLS → collect network traffic → exported activity testing.
Quick Start
One-command setup
bash scripts/setup.sh
Or step-by-step
# 1. Start MobSF
docker compose up -d
# 2. Get the API key
docker logs mobsf 2>&1 | grep "REST API Key"
# 3. Install the MCP server
pip install -e .
# 4. Configure
cp .env.example .env
# Edit .env with your API key
# 5. Use with your MCP client
mobsf-mcp
Connecting to Your LLM
Once MobSF is running and the MCP server is installed, configure your MCP client to load the tools.
Claude Desktop
Add to ~/.claude/claude_desktop_config.json:
{
"mcpServers": {
"mobsf": {
"command": "mobsf-mcp",
"env": {
"MOBSF_URL": "http://localhost:8000",
"MOBSF_API_KEY": "your_api_key_here"
}
}
}
}
Restart Claude Desktop. The 23 tools appear automatically.
Open WebUI + Ollama
See docs/ollama-setup.md for connecting local LLMs via mcpo bridge.
Any MCP Client
The server speaks stdio MCP. Point any MCP-compatible client at the mobsf-mcp command with MOBSF_URL and MOBSF_API_KEY environment variables. No additional configuration needed.
# Test standalone
MOBSF_URL=http://localhost:8000 MOBSF_API_KEY=your_key mobsf-mcp
Available Tools (23)
Static Analysis (8)
| Tool | Description |
|---|---|
upload_apk |
Upload an APK file for analysis |
scan_apk |
Trigger static analysis (decompile, manifest, code, secrets) |
get_report |
Get detailed security findings (filterable by section) |
get_security_scorecard |
High-level security posture score |
view_source_file |
Inspect specific decompiled source files |
list_scans |
List all previous scans |
search_scans |
Search scans by name/package/hash |
delete_scan |
Remove a scan |
Dynamic Analysis (9)
| Tool | Description |
|---|---|
list_dynamic_apps |
List apps ready for dynamic testing |
start_dynamic_analysis |
Install & launch app on emulator |
stop_dynamic_analysis |
Stop analysis & collect results |
get_dynamic_report |
Get runtime analysis findings |
get_logcat |
Get filtered logcat output |
run_adb_command |
Execute ADB commands on device |
test_exported_activities |
Test for unauthorized activity access |
launch_activity |
Launch a specific activity |
run_tls_tests |
Test TLS/SSL security |
Frida Instrumentation (6)
| Tool | Description |
|---|---|
frida_instrument |
Run hooks (SSL bypass, root bypass, custom scripts) |
frida_monitor_apis |
Monitor sensitive API calls at runtime |
frida_get_logs |
Get Frida hook output |
frida_list_scripts |
List available built-in scripts |
frida_get_script_code |
View script source code |
frida_get_dependencies |
Install Frida dependencies on device |
Dynamic Analysis Setup
Dynamic analysis requires an Android emulator with root access (for writable /system) connected to MobSF.
Note on API level: MobSF v4.5.1 uses Frida 17.15.3 which has a known compatibility issue with Android 13 (API 33) -
frida.attach()crashes with aTypeError. For full DAST including Frida instrumentation, use an Android 11 (API 30) emulator. ADB commands, exported activity testing, TLS tests, and logcat all work fine on API 33. This will be resolved when MobSF updates its Frida dependency.
Emulator Setup
Two options:
Option A: Docker Android Emulator (simplest)
Use docker-android - a pre-built Docker image with Android emulator, SDK, and ADB ready to go. No local SDK install needed.
docker run -d --privileged -p 6080:6080 -p 5554:5554 -p 5555:5555 \
-e DEVICE="pixel_6" -e EMULATOR_ARGS="-writable-system" \
budtmo/docker-android:emulator_13.0
# Wait for boot, then root it
adb connect localhost:5555
adb root && adb remount
Then set ANALYZER_IDENTIFIER=localhost:5555 in docker-compose.yml (instead of emulator-5554).
Option B: Manual AVD (more control)
# Install Android SDK command-line tools, then:
sdkmanager "system-images;android-33;google_apis;x86_64" "platforms;android-33" "platform-tools"
avdmanager create avd -n mobsf -k "system-images;android-33;google_apis;x86_64" -d pixel_6 -f
# Start with writable system
emulator -avd mobsf -no-window -no-audio -no-snapshot \
-writable-system -gpu swiftshader_indirect -memory 2048 &
# Wait for boot
adb wait-for-device
adb root && adb remount
adb shell 'while [[ -z $(getprop sys.boot_completed) ]]; do sleep 5; done'
Docker ↔ Emulator Wiring
This project's docker-compose.yml is pre-configured:
network_mode: host- Docker shares the host network, so ADB just worksMOBSF_PLATFORM=host- disables Docker Desktop hostname translationANALYZER_IDENTIFIER=emulator-5554- points MobSF to the emulator./mobsf-data/.android/- ADB keys mounted for authorization- Entrypoint wrapper patches MobSF to support API 33+
Important: Kill the host ADB server before starting MobSF Docker to avoid ADB conflicts:
adb kill-server && docker compose up -d
Example Agent Interaction
User: Analyze the security of ./target.apk
Agent: I'll upload and scan this APK for you.
[calls upload_apk("./target.apk")]
[calls scan_apk("a1b2c3d4...")]
[calls get_security_scorecard("a1b2c3d4...")]
The app has a security score of 28/100 (High Risk):
- 6 high-severity manifest issues (StrandHogg 2.0, debuggable)
- 7 exported components without permission checks
- Janus vulnerability (v1 signature only)
- 23 hardcoded secrets found
- 3 privacy trackers
- Debug enabled, backup allowed
Want me to start dynamic analysis and test the exported activities?
File Structure
/home/kali/labs/MCP/
├── src/
│ ├── server.py ← MCP server (23 tools)
│ └── mobsf_client.py ← MobSF REST API client
├── scripts/
│ ├── setup.sh ← One-command setup
│ └── test_pipeline.py ← End-to-end test suite
├── docs/
│ └── ollama-setup.md ← LLM integration guide
├── mobsf-data/ ← Docker volume (persistent)
│ ├── .android/adbkey ← ADB authorization keys
│ ├── entrypoint-wrapper.sh ← API cap patch
│ ├── config.py / db.sqlite3 / secret
│ └── downloads/ ← Scan artifacts
├── tests/
│ └── test.apk ← InsecureBankv2 (test target)
├── docker-compose.yml ← MobSF container config
├── pyproject.toml ← Python package config
├── .env.example ← Environment template
└── README.md
Development
pip install -e .
python3 scripts/test_pipeline.py
License
MIT © plur1bu5
<p align="center"> <sub>Built with ❤️ by <a href="https://github.com/plur1bu5">plur1bu5</a> - if this helps you hack smarter, drop a ⭐</sub> </p>
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.