Agentic Detection Lookups

Agentic Detection Lookups

Machine-readable detection lookups for SIEM enrichment and AI agents. MCP-native.

Stop regex-matching 200+ binaries. Enrich in one match() call.
Feed it to your SIEM, your SOAR, your agent, or your LLM.

What is this?

A collection of structured CSV lookup files purpose-built for:

• SIEM enrichment — one match()/lookup/join replaces entire rule categories
• AI agent tooling — MCP server included, agents query detection context in real-time
• Detection automation — consistent schema, CI-updated, deploy-ready

Lookup Files

File	Entries	OS	Description
lolbas_binaries.csv	234	Windows	Living Off The Land Binaries and Scripts — risk-scored, categorized, MITRE-mapped
gtfobins.csv	477	Linux	GTFOBins Unix binaries — shell escape, priv-esc, file ops, MITRE-mapped
parent_child_baselines.csv	98	Both	Expected/suspicious process parent→child relationships for Windows and Linux

Schema Contract

Every lookup file follows:

1. First column = match key (the field you join on)
2. Always includes risk or risk_if_unexpected column
3. Always includes MITRE ATT&CK technique mapping
4. No nested data — flat columns, pipe-delimited for multi-value
5. UTF-8, no BOM, Unix line endings, header row always present

Quick Start

SIEM (copy-paste)

CrowdStrike NG-SIEM:

#event_simpleName=ProcessRollup2
| binary := lower(FileName)
| match(file="lolbas_binaries.csv", field=binary, column=filename, include=[categories, mitre_ids, risk])
| risk="high"

Splunk:

index=crowdstrike event_simpleName=ProcessRollup2
| rex field=FileName "(?<binary>[^\\\\]+)$"
| lookup lolbas_binaries.csv filename AS binary OUTPUT categories mitre_ids risk
| where risk="high"

Elastic (ES|QL):

FROM logs-endpoint.events.process-*
| WHERE event.action == "start"
| ENRICH lolbas-policy ON process.name = filename WITH categories, risk
| WHERE risk == "high"

Microsoft Sentinel:

DeviceProcessEvents
| extend binary = tolower(FileName)
| join kind=inner (_GetWatchlist('lolbas_binaries')) on $left.binary == $right.filename
| where risk == "high"

See queries/ for full query libraries per platform.

MCP Server (AI agents)

{
  "servers": {
    "detection-lookups": {
      "type": "stdio",
      "command": "python",
      "args": ["-m", "mcp_server"],
      "cwd": "/path/to/agentic-detection-lookups"
    }
  }
}

Then your agent can:

→ detection_lookup_binary("certutil.exe")
← {source: "lolbas", risk: "medium", categories: ["Download"], mitre_ids: ["T1105"]}

→ detection_lookup_binary("python")
← {source: "gtfobins", risk: "high", categories: ["shell", "reverse-shell", ...], mitre_ids: ["T1059"]}

→ detection_check_parent_child("winword.exe", "cmd.exe")
← {expected: false, risk_if_unexpected: "critical", mitre_id: "T1204.002"}

MCP Tools

Tool	Input	Output
detection_lookup_binary	filename	Risk, categories, MITRE IDs, source (lolbas/gtfobins)
detection_check_parent_child	parent, child, os_filter	Expected/suspicious, risk level, triage guidance
detection_list_by_category	category, limit, offset	Paginated binaries in that abuse category (cross-platform)
detection_list_by_mitre	technique_id, limit, offset	Paginated binaries mapped to that technique (cross-platform)
detection_search	query, limit	Matches across all lookup data with total/has_more
detection_list_lookups	—	All files with row counts and columns

Data Sources

Lookup	Source	Update Frequency
LOLBAS binaries	LOLBAS Project	Weekly (automated)

Installation

Prerequisites

• Python 3.10+
• VS Code with GitHub Copilot (for MCP integration)

Install

git clone https://github.com/detection-forge/agentic-detection-lookups.git
cd agentic-detection-lookups
python -m venv .venv
# Windows:
.venv\Scripts\activate
# Linux/macOS:
source .venv/bin/activate
pip install -e .

Configure MCP Client (VS Code)

Add to your VS Code User settings (Ctrl+Shift+P → "Preferences: Open User Settings (JSON)") or ~/.vscode/mcp.json:

{
  "servers": {
    "detection-lookups": {
      "type": "stdio",
      "command": "/absolute/path/to/.venv/bin/python",
      "args": ["-m", "mcp_server"],
      "cwd": "/absolute/path/to/agentic-detection-lookups"
    }
  }
}

Windows example:

{
  "servers": {
    "detection-lookups": {
      "type": "stdio",
      "command": "C:\\Code\\.venv\\Scripts\\python.exe",
      "args": ["-m", "mcp_server"],
      "cwd": "C:\\Code\\agentic-detection-lookups"
    }
  }
}

Reload VS Code: Ctrl+Shift+P → "Reload Window"

Verify

In Copilot Chat (Agent mode):

Is certutil.exe a LOLBAS binary?

✅ Returns risk, categories, and MITRE mappings = working!

Run standalone (CLI)

detection-lookups

This starts the MCP server on stdio transport (useful for piping JSON-RPC or connecting other MCP clients).

Upload to your SIEM

• CrowdStrike NG-SIEM: Upload via API or UI (Settings → Lookup Files)
• Splunk: Settings → Lookups → Lookup table files → Add new
• Elastic: Create enrich index + ingest pipeline
• Sentinel: Configuration → Watchlist → Add new

Project Structure

agentic-detection-lookups/
├── lookups/                    # The data (CSV files)
│   ├── lolbas_binaries.csv
│   ├── gtfobins.csv
│   └── parent_child_baselines.csv
├── queries/                    # Copy-paste detection queries
│   ├── crowdstrike_ngsiem.md
│   ├── splunk.md
│   ├── elastic.md
│   └── microsoft_sentinel.md
├── mcp_server/                 # MCP server for AI agents
│   ├── server.py
│   └── __init__.py
├── scripts/                    # Update/maintenance scripts
├── LICENSE                     # Apache 2.0
├── NOTICE
└── pyproject.toml

Contributing

PRs welcome. See CONTRIBUTING.md for guidelines.

To add a new lookup file:

1. Follow the schema contract (match key first, include risk + MITRE columns)
2. Include at least one query example per SIEM platform
3. Add a tool to the MCP server

License

Apache 2.0 — See LICENSE and NOTICE.

---

Built by Gene Kazimiarovich | Part of Detection Forge