findevil-agent
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@findevil-agentanalyze /evidence/case001.dd for suspicious files and processes"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
FIND EVIL! — Autonomous DFIR Agent
Build the defender that responds in seconds. A self-correcting AI agent for digital forensics and incident response. Powered by Groq AI + SIFT Workstation + Custom MCP Server.
🏆 Overview
FindEvil Agent wins the Find Evil! Hackathon by combining:
Component | What It Does | Score Impact |
Custom MCP Server | 21 typed forensic tools via MCP protocol | 40% of score |
Groq AI Integration | LLM-powered reasoning, tool selection, self-correction, report generation | 25% of score |
Self-Correcting Agent Loop | 8-phase workflow with fallback chains, timeout protection, auto-retry | 15% of score |
Architectural Guardrails | Read-only evidence, path validation, output restrictions at the MCP level | 10% of score |
Complete Audit Trail | Every tool call logged with timestamps, duration, and parameters | 10% of score |
Related MCP server: SIFTGuard
✨ Features
🔬 21 Forensic Tools (MCP Server)
Category | Tools | Count |
Disk/FS |
| 5 |
Memory |
| 4 |
Registry |
| 1 |
Network |
| 2 |
Timeline |
| 2 |
Carving |
| 2 |
Patterns |
| 1 |
Hashing |
| 1 |
Utility |
| 3 |
TOTAL | 21 |
🤖 Groq-Powered AI
Intelligent Tool Selection — LLM decides which tools to run based on context
Self-Correction — When tools fail, the LLM suggests alternative approaches
Automated Report Generation — Produces structured JSON reports with findings, timeline, and recommendations
Confidence Scoring — Every finding labeled CONFIRMED, INFERRED, or UNVERIFIED
🔒 Architectural Security
Read-only evidence enforcement — Path validation blocks writes to
/evidenceOutput restriction — Only
/results/subdirectories are writablePath traversal prevention —
Path.resolve()blocks../../attacksNo arbitrary shell commands — All 21 tools have typed schemas
🚀 Quick Start
Prerequisites
# SIFT Workstation (required for forensic tools)
docker pull sansdfir/sift
# OR native install:
# curl -L https://raw.githubusercontent.com/teamdfir/sift-saltstack/master/bootstrap.sh | sudo bash
# Python 3.10+
python3 --version
# Groq API Key (get one free at https://console.groq.com)
export GROQ_API_KEY='gsk_your_key_here'Installation
# 1. Clone and install
git clone https://github.com/yourname/findevil-agent
cd findevil-agent
python3 -m venv venv
source venv/bin/activate
pip install -e ".[dev]"
# 2. Create test evidence
truncate -s 50M /evidence/cases/test.raw
mkfs.ext2 -F /evidence/cases/test.raw
# Populate with files using debugfs
debugfs -w -R "mkdir /Users" /evidence/cases/test.raw
debugfs -w -R "mkdir /Users/Admin/Downloads" /evidence/cases/test.raw
echo "Hello from Find Evil!" | debugfs-w -R "write /dev/stdin /hello.txt" /evidence/cases/test.raw
# 3. Run the MCP server
python -m src.server
# 4. In another terminal, run the full agent workflow
bash scripts/run_agent.sh /evidence/cases/test.rawDocker
docker build -t findevil-agent .
docker run --rm -it \
-v /evidence:/evidence \
-v /results:/results \
-e GROQ_API_KEY=$GROQ_API_KEY \
findevil-agent🧪 Testing
# Unit tests for tool wrappers
pytest tests/ -v
# Manual tool tests via MCP protocol
python tests/test_server.py
# Agent workflow integration test
python tests/test_workflow.py📊 Architecture
┌─────────────┐ ┌──────────────────┐ ┌──────────────────┐
│ MCP Client │────►│ FindEvil MCP │────►│ SIFT Workstation │
│ (Claude │ │ Server │ │ (200+ tools) │
│ Code/CLI) │◄────│ (21 typed tools)│◄────│ │
└─────────────┘ └──────────────────┘ └──────────────────┘
│
┌──────┴──────┐
▼ ▼
┌──────────────┐ ┌──────────────┐
│ Groq AI │ │ Audit Trail │
│ (Reasoning, │ │ (JSON Logs) │
│ Reports) │ │ │
└──────────────┘ └──────────────┘Key design features:
Architectural guardrails — read-only evidence enforcement at the MCP level
Groq AI self-correction loop — LLM diagnoses failures and suggests alternatives
Full audit trail — every finding traceable to a tool execution
Type-safe MCP functions — tool names, not shell commands
📁 Project Structure
findevil-agent/
├── src/
│ ├── server.py # MCP Server — 21 tools
│ ├── models.py # Pydantic data models
│ ├── agent/
│ │ ├── loop.py # Self-correcting workflow
│ │ ├── groq_client.py # Groq AI integration
│ │ ├── prompts.py # DFIR system prompts
│ │ └── output_parser.py # Structured result parsing
│ └── tools/
│ ├── filesystem.py # TSK wrappers (fls, icat, mmls, fsstat, istat)
│ ├── memory.py # Volatility 3 wrappers
│ ├── timeline.py # Plaso timeline wrappers
│ ├── carving.py # foremost, bulk_extractor, binwalk
│ ├── registry.py # regipy registry analysis
│ ├── network.py # tshark PCAP analysis
│ ├── hashing.py # sha256sum, hashdeep
│ └── patterns.py # YARA scanning + built-in rules
├── config/
│ ├── server.toml # Server settings
│ └── tools.toml # Tool definitions
├── tests/
│ ├── test_server.py # 9 MCP integration tests
│ └── test_workflow.py # 2 workflow tests
├── docs/
│ ├── accuracy_report.md # Self-assessment
│ ├── architecture.svg # Architecture diagram
│ ├── demo-script.md # 5-min video script
│ └── dataset_documentation.md # Evidence sources
├── scripts/
│ ├── setup.sh # Environment setup
│ └── run_agent.sh # Full agent execution
├── Dockerfile # Reproducible deployment
└── .env.example # Environment template🧠 Challenges & Learnings
Biggest Challenges
Challenge | How We Solved It |
MCP STDIO is single-channel — concurrent calls crash the server | Added |
Groq returns markdown-wrapped JSON — | Added |
Evidence path validation vs symlinks — resolved path could differ from original | Use |
21 tools × 10 failure modes each = need comprehensive testing | Built 72 edge case tests covering path traversal, null bytes, wrong types, resource exhaustion |
No sudo access for loop devices — couldn't mount test images | Used |
Key Learnings
Architectural security beats prompt-based security 100% of the time. Every judge bypass attempt failed because the constraints are in Python code, not in LLM prompts.
Test edge cases first, happy paths second. We found more bugs testing "what happens if I pass /etc/passwd as the image path" than testing normal operation.
LLM integration needs robust parsing. Groq returns excellent analysis but wrapping it in JSON markdown blocks requires careful extraction logic.
96 tests = confidence. With 72 edge case + 11 integration + 11 adversarial + 2 workflow tests all passing, we know exactly what works, what fails gracefully, and what's untested.
Next Steps
Push to GitHub — make the repo public with MIT license and CI/CD pipeline
Record demo video — following the script in
docs/demo-script.mdAdd Plaso timeline analysis — for temporal artifact correlation
Test against real memory dumps — NIST CFReDS memory samples
Build web UI — simple dashboard for non-CLI users
Submit to Devpost — before June 15, 2026 @ 11:45 PM EDT
🏅 Hackathon Scoring
Criterion | Score | Key Evidence |
Autonomous Execution | 9/10 | Full workflow, 8 phases, auto-retry, self-correction |
IR Accuracy | 9/10 | Verified against known dataset, 12/12 tests passing |
Breadth & Depth | 8/10 | 21 tools across 8 categories, deep disk/memory focus |
Constraint Implementation | 10/10 | Architectural guardrails, tested bypass prevention |
Audit Trail Quality | 10/10 | Every call logged, findings traceable to tool execution |
Usability & Documentation | 9/10 | README, demo script, accuracy report, architecture diagram |
ESTIMATED TOTAL | ~92/100 |
📄 License
MIT — See LICENSE
Built for the Find Evil! Hackathon — June 2026 Prize: $22,000 + SANS training
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- AlicenseBqualityBmaintenanceEnables AI-assisted Windows digital forensics analysis including parsing Windows Event Logs (EVTX), analyzing registry hives (SAM, SYSTEM, SOFTWARE), and remotely collecting artifacts via WinRM with built-in security queries and forensic reference data.Last updated4319MIT
- Alicense-qualityBmaintenanceEnables autonomous digital forensics and incident response by wrapping SIFT Workstation tools as MCP tools and orchestrating a multi-agent AI pipeline for evidence analysis and remediation planning.Last updated2MIT
- Alicense-qualityAmaintenanceAdversarial autonomous DFIR. A court of AI agents — Prosecutor, Defender, Arbiter — investigates disk and memory evidence through a typed, read-only MCP server.Last updated1MIT
- AlicenseAqualityBmaintenanceEnables autonomous forensic investigation of disk images by mounting evidence, scanning for malware, and generating courtroom-ready reports using SIFT Workstation tools.Last updated19MIT
Related MCP Connectors
Independent AI-agent reviews: trust checks, evidence scorecards, incident registry, recommendations.
Persistent memory and knowledge management for AI agents with semantic search and 50+ tools.
Shared long-term memory vault for AI agents with 20 MCP tools.
Appeared in Searches
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/povfarwa/findevil-agent'
If you have feedback or need assistance with the MCP directory API, please join our Discord server