MCP Server Fuzzer

A comprehensive super-aggressive CLI-based fuzzing tool for MCP servers

Multi-protocol support • Two-phase fuzzing • Built-in safety • Rich reporting • async runtime and async fuzzing of mcp tools

Documentation • Quick Start • Examples • Configuration

What is MCP Server Fuzzer?

MCP Server Fuzzer is a comprehensive fuzzing tool designed specifically for testing Model Context Protocol (MCP) servers. It supports both tool argument fuzzing and protocol type fuzzing across multiple transport protocols.

Key Promise

If your server conforms to the MCP schema, this tool will fuzz it effectively and safely.

Why Choose MCP Server Fuzzer?

Safety First: Built-in safety system prevents dangerous operations
High Performance: Asynchronous execution with configurable concurrency
Beautiful Output: Rich, colorized terminal output with detailed reporting
Flexible Configuration: CLI args, YAML configs, environment variables
Comprehensive Reporting: Multiple output formats (JSON, CSV, HTML, Markdown)
Production Ready: PATH shims, sandbox defaults, and CI-friendly controls
Intelligent Testing: Hypothesis-based data generation with custom strategies
More Than Conformance: Goes beyond the checks in modelcontextprotocol/conformance with fuzzing, reporting, and safety tooling

Extensibility for Contributors

MCP Server Fuzzer is designed for easy extension while keeping CLI usage simple:

Custom Transports: Add support for new protocols via config or self-registration (see docs/transport/custom-transports.md).
Pluggable Safety: Swap safety providers for custom filtering rules.
Injectable Components: Advanced users can inject custom clients/reporters for testing or plugins.

The modularity improvements (dependency injection, registries) make it maintainer-friendly without complicating the core CLI experience.

Quick Start

Installation

Requires Python 3.10+ (editable installs from source also need a modern pip).

# Install from PyPI
pip install mcp-fuzzer

# Or install from source (includes MCP spec submodule)
git clone --recursive https://github.com/Agent-Hellboy/mcp-server-fuzzer.git
cd mcp-server-fuzzer
# If you already cloned without submodules, run:
git submodule update --init --recursive
pip install -e .

Docker Installation

The easiest way to use MCP Server Fuzzer is via Docker:

# Build the Docker image
docker build -t mcp-fuzzer:latest .

# Or pull the published image
# docker pull princekrroshan01/mcp-fuzzer:latest

The container ships with mcp-fuzzer as the entrypoint, so you pass CLI args after the image name. Use /output for reports and mount any server/config inputs you need.

# Show CLI help
docker run --rm mcp-fuzzer:latest --help

# Example: store reports on the host
docker run --rm -v $(pwd)/reports:/output mcp-fuzzer:latest \
  --mode tools --protocol http --endpoint http://localhost:8000 \
  --output-dir /output

Required mounts (stdio/config workflows):

/output: writeable reports directory
/servers: read-only server code/executables for stdio
/config: read-only config directory

Basic Usage

Set up your MCP server (HTTP, SSE, or Stdio)
Run basic fuzzing:

Using Docker:

# Fuzz HTTP server (container acts as client)
docker run --rm -it --network host \
  -v $(pwd)/reports:/output \
  mcp-fuzzer:latest \
  --mode tools --protocol http --endpoint http://localhost:8000

# Fuzz stdio server (server runs in containerized environment)
docker run --rm -it \
  -v $(pwd)/servers:/servers:ro \
  -v $(pwd)/reports:/output \
  mcp-fuzzer:latest \
  --mode tools --protocol stdio --endpoint "node /servers/my-server.js stdio"

Using Local Installation:

# Fuzz tools on an HTTP server
mcp-fuzzer --mode tools --protocol http --endpoint http://localhost:8000

# Fuzz protocol types on an SSE server
mcp-fuzzer --mode protocol --protocol-type InitializeRequest --protocol sse --endpoint http://localhost:8000/sse

Advanced Usage

# Two-phase fuzzing (realistic + aggressive)
mcp-fuzzer --mode all --phase both --protocol http --endpoint http://localhost:8000

# With safety system enabled
mcp-fuzzer --mode tools --enable-safety-system --safety-report

# Export results to multiple formats
mcp-fuzzer --mode tools --export-csv results.csv --export-html results.html

# Use configuration file
mcp-fuzzer --config my-config.yaml

Examples

HTTP Server Fuzzing

# Basic HTTP fuzzing
mcp-fuzzer --mode tools --protocol http --endpoint http://localhost:8000 --runs 50

# With authentication
mcp-fuzzer --mode tools --protocol http --endpoint https://api.example.com \
           --auth-config auth.json --runs 100

SSE Server Fuzzing

# SSE protocol fuzzing
mcp-fuzzer --mode protocol --protocol-type InitializeRequest --protocol sse --endpoint http://localhost:8080/sse \
           --runs-per-type 25 --verbose

Stdio Server Fuzzing

Using Docker (Recommended for Isolation):

# Server runs in containerized environment for safety
docker run --rm -it \
  -v $(pwd)/servers:/servers:ro \
  -v $(pwd)/reports:/output \
  mcp-fuzzer:latest \
  --mode tools --protocol stdio --endpoint "python /servers/my_server.py" \
  --enable-safety-system --fs-root /tmp/safe \
  --output-dir /output

# Using docker-compose (easier configuration)
docker-compose run --rm fuzzer \
  --mode tools --protocol stdio --endpoint "node /servers/my-server.js stdio" \
  --runs 50 --output-dir /output

Using Local Installation:

# Local server testing
mcp-fuzzer --mode tools --protocol stdio --endpoint "python my_server.py" \
           --enable-safety-system --fs-root /tmp/safe

Configuration File Usage

# config.yaml
mode: tools
protocol: stdio
endpoint: "python dev_server.py"
runs: 10
phase: realistic

# Optional output configuration
output:
  directory: "reports"
  format: "json"
  types:
    - "fuzzing_results"
    - "safety_summary"

mcp-fuzzer --config config.yaml

Docker Usage

MCP Server Fuzzer can be run in a Docker container, providing isolation and easy deployment. This is especially useful for:

Stdio Servers: Run servers in a containerized environment for better isolation and safety
HTTP/SSE Servers: Container acts as the MCP client (server can run anywhere)
CI/CD Pipelines: Consistent testing environment across different systems

Quick Start with Docker

# Build the image
docker build -t mcp-fuzzer:latest .

# Fuzz HTTP server (server can be on host or remote)
docker run --rm -it --network host \
  -v $(pwd)/reports:/output \
  mcp-fuzzer:latest \
  --mode tools --protocol http --endpoint http://localhost:8000 --output-dir /output

# Fuzz stdio server (server code mounted from host)
docker run --rm -it \
  -v $(pwd)/servers:/servers:ro \
  -v $(pwd)/reports:/output \
  mcp-fuzzer:latest \
  --mode tools --protocol stdio --endpoint "python /servers/my_server.py" --output-dir /output

Docker Releases

Docker images are published automatically on every GitHub Release (tagged v*) via CI. The published image is:

docker pull princekrroshan01/mcp-fuzzer:latest

Note: The runtime image includes curl and ca-certificates so stdio servers can fetch HTTPS resources (e.g., schemas, tokens, metadata) without bundling extra tools. If your servers never make outbound HTTPS calls, you can remove them.

Using Docker Compose

For easier configuration and management, use docker-compose.yml:

# Set environment variables (optional)
export SERVER_PATH=./servers
export CONFIG_PATH=./examples/config
export MCP_SPEC_SCHEMA_VERSION=2025-06-18

# Run fuzzing (stdio server)
docker-compose run --rm fuzzer \
  --mode tools \
  --protocol stdio \
  --endpoint "node /servers/my-server.js stdio" \
  --runs 50 \
  --output-dir /output

# For HTTP servers (macOS/Windows - uses host.docker.internal)
docker-compose run --rm fuzzer \
  --mode tools \
  --protocol http \
  --endpoint http://host.docker.internal:8000 \
  --runs 50 \
  --output-dir /output

# For HTTP servers on Linux (use host network)
docker-compose -f docker-compose.host-network.yml run --rm fuzzer \
  --mode tools \
  --protocol http \
  --endpoint http://localhost:8000 \
  --runs 50 \
  --output-dir /output

# Production-style (no TTY/stdin)
docker-compose -f docker-compose.prod.yml run --rm fuzzer \
  --mode tools \
  --protocol stdio \
  --endpoint "node /servers/my-server.js stdio" \
  --runs 50 \
  --output-dir /output

Docker Volume Mounts

/output: Mount your reports directory here (e.g., -v $(pwd)/reports:/output)
/servers: Mount server code/executables for stdio servers (read-only recommended)
/config: Mount custom configuration files if needed

Network Configuration

HTTP/SSE Servers: Network access required. Linux: prefer --network host so localhost works. Docker Desktop (macOS/Windows): use host.docker.internal since host networking is limited. If neither works, use the host IP.
Stdio Servers: No network needed - server runs as subprocess in container

Example: Fuzzing a Node.js Stdio Server

# 1. Prepare your server
mkdir -p servers
cp my-mcp-server.js servers/

# 2. Run fuzzer in Docker
docker run --rm -it \
  -v $(pwd)/servers:/servers:ro \
  -v $(pwd)/reports:/output \
  mcp-fuzzer:latest \
  --mode all \
  --protocol stdio \
  --endpoint "node /servers/my-mcp-server.js stdio" \
  --runs 100 \
  --enable-safety-system \
  --output-dir /output \
  --export-json /output/results.json

Example: Fuzzing an HTTP Server

# Server runs on host at localhost:8000
# Container connects to it as client

docker run --rm -it --network host \
  -v $(pwd)/reports:/output \
  mcp-fuzzer:latest \
  --mode tools \
  --protocol http \
  --endpoint http://localhost:8000 \
  --runs 50 \
  --output-dir /output

Security Considerations

The Docker container runs as non-root user (UID 1000) for improved security
Stdio servers run in isolated container environment
Use read-only mounts (:ro) for server code when possible
Reports are written to mounted volume, not inside container

Configuration

Configuration Methods (in order of precedence)

Command-line arguments (highest precedence)
Configuration files (YAML)
Environment variables (lowest precedence)

Environment Variables

# Core settings
export MCP_FUZZER_TIMEOUT=60.0
export MCP_FUZZER_LOG_LEVEL=DEBUG

# Safety settings
export MCP_FUZZER_SAFETY_ENABLED=true
export MCP_FUZZER_FS_ROOT=/tmp/safe

# Authentication
export MCP_API_KEY="your-api-key"
export MCP_USERNAME="your-username"
export MCP_PASSWORD="your-password"

Performance Tuning

# High concurrency for fast networks
mcp-fuzzer --process-max-concurrency 20 --watchdog-check-interval 0.5

# Conservative settings for slow/unreliable servers
mcp-fuzzer --timeout 120 --process-retry-count 5 --process-retry-delay 2.0

Key Features

Feature	Description
Two-Phase Fuzzing	Realistic testing + aggressive security testing
Multi-Protocol Support	HTTP, SSE, Stdio, and StreamableHTTP transports
Built-in Safety	Pattern-based filtering, sandboxing, and PATH shims
Intelligent Testing	Hypothesis-based data generation with custom strategies
Rich Reporting	Detailed output with exception tracking and safety reports
Multiple Output Formats	JSON, CSV, HTML, Markdown, and XML export options
Flexible Configuration	CLI args, YAML configs, environment variables
Asynchronous Execution	Efficient concurrent fuzzing with configurable limits
Comprehensive Monitoring	Process watchdog, timeout handling, and resource management
Authentication Support	API keys, basic auth, OAuth, and custom providers
Performance Metrics	Built-in benchmarking and performance analysis
Schema Validation	Automatic MCP protocol compliance checking

Performance

Concurrent Operations: Up to 20 simultaneous fuzzing tasks
Memory Efficient: Streaming responses and configurable resource limits
Fast Execution: Optimized async I/O and connection pooling
Scalable: Configurable timeouts and retry mechanisms

Architecture

The system is built with a modular architecture:

CLI Layer: User interface and argument handling
Transport Layer: Protocol abstraction (HTTP/SSE/Stdio)
Fuzzing Engine: Test orchestration and execution
Strategy System: Data generation (realistic + aggressive)
Safety System: Core filter + SystemBlocker PATH shim; safe mock responses
Runtime: Fully async ProcessManager + ProcessWatchdog
Authentication: Multiple auth provider support
Reporting: FuzzerReporter, Console/JSON/Text formatters, SafetyReporter

Runtime Watchdog Overview

The watchdog supervises processes registered through ProcessManager, combining hang detection, signal dispatch, and registry-driven cleanup. For a deeper dive into lifecycle events, custom signal strategies, and registry wiring, see the runtime management guide.

Understanding the Design Patterns

For developers (beginners to intermediate) who want to understand the design patterns used throughout the codebase, please refer to our comprehensive Design Pattern Review. This document provides:

Module-by-module pattern analysis
Design pattern fit scores and recommendations
Modularity observations and improvement suggestions
Complete pattern map for every module in the codebase

This is especially helpful if you're:

Learning about design patterns in real-world applications
Planning to contribute to the project
Wanting to understand the architectural decisions
Looking for areas to improve or extend

Troubleshooting

Common Issues

Connection Timeout

# Increase timeout for slow servers
mcp-fuzzer --timeout 120 --endpoint http://slow-server.com

Authentication Errors

# Check auth configuration
mcp-fuzzer --check-env
mcp-fuzzer --validate-config config.yaml

Memory Issues

# Reduce concurrency for memory-constrained environments
mcp-fuzzer --process-max-concurrency 2 --runs 25

Permission Errors

# Run with appropriate permissions or use safety system
mcp-fuzzer --enable-safety-system --fs-root /tmp/safe

Debug Mode

# Enable verbose logging
mcp-fuzzer --verbose --log-level DEBUG

# Check environment
mcp-fuzzer --check-env

Community & Support

Documentation: Full Documentation
Issues: GitHub Issues
Discussions: GitHub Discussions

Contributing

We welcome contributions! Please see our Contributing Guide for details.

Quick Start for Contributors:

git clone --recursive https://github.com/Agent-Hellboy/mcp-server-fuzzer.git
cd mcp-server-fuzzer
# If you already cloned without submodules, run:
git submodule update --init --recursive
pip install -e .[dev]
pytest tests/

License

This project is licensed under the MIT License - see the LICENSE file for details.

Disclaimer

This tool is designed for testing and security research purposes only.

Always use in controlled environments
Ensure you have explicit permission to test target systems
The safety system provides protection but should not be relied upon as the sole security measure
Use at your own risk

Funding & Support

If you find this project helpful, please consider supporting its development:

Ways to support:

⭐ Star the repository - helps others discover the project
🐛 Report issues - help improve the tool
💡 Suggest features - contribute ideas for new functionality
💰 Sponsor on GitHub - directly support ongoing development
📖 Share the documentation - help others learn about MCP fuzzing

Your support helps maintain and improve this tool for the MCP community!

Made with love for the MCP community

Star us on GitHub • Read the Docs

Name		Name	Last commit message	Last commit date
Latest commit History 308 Commits
.github		.github
docker		docker
docs		docs
examples		examples
images		images
mcp_fuzzer		mcp_fuzzer
schemas		schemas
tests		tests
.dockerignore		.dockerignore
.gitignore		.gitignore
.gitmodules		.gitmodules
.pre-commit-config.yaml		.pre-commit-config.yaml
Dockerfile		Dockerfile
LICENSE		LICENSE
README.md		README.md
docker-compose.host-network.yml		docker-compose.host-network.yml
docker-compose.prod.yml		docker-compose.prod.yml
docker-compose.yml		docker-compose.yml
icon.png		icon.png
mkdocs.yml		mkdocs.yml
pyproject.toml		pyproject.toml
requirements.txt		requirements.txt
tox.ini		tox.ini
uv.lock		uv.lock

Uh oh!

License

Agent-Hellboy/mcp-server-fuzzer

Folders and files

Latest commit

History

Repository files navigation

MCP Server Fuzzer

What is MCP Server Fuzzer?

Key Promise

Why Choose MCP Server Fuzzer?

Extensibility for Contributors

Quick Start

Installation

Docker Installation

Basic Usage

Advanced Usage

Examples

HTTP Server Fuzzing

SSE Server Fuzzing

Stdio Server Fuzzing

Configuration File Usage

Docker Usage

Quick Start with Docker

Docker Releases

Using Docker Compose

Docker Volume Mounts

Network Configuration

Example: Fuzzing a Node.js Stdio Server

Example: Fuzzing an HTTP Server

Security Considerations

Configuration

Configuration Methods (in order of precedence)

Environment Variables

Performance Tuning

Key Features

Performance

Architecture

Runtime Watchdog Overview

Understanding the Design Patterns

Troubleshooting

Common Issues

Debug Mode

Community & Support

Contributing

License

Disclaimer

Funding & Support

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 18

Sponsor this project

Uh oh!

Packages 0

Contributors 3

Uh oh!

Languages

Packages