Unmasking the Moon: Comparing LunaStealer Samples with MalChela and Claude

As one tends to do on Saturday mornings with coffee in hand, I was reviewing two samples that were attributed to the LunaStealer / LunaGrabber family. Originally I was validating that tiquery was working with the MCP configuration, however what started as a quick TI check turned into a full static analysis session — and it gave me a good opportunity to put the MalChela MCP integration through its paces in a real workflow. This post walks through how that investigation unfolded, what the pivot points were, and what we found at the bottom of the rabbit hole.


The Setup

If you haven’t seen the MalChela MCP plugin before, the short version is this: MalChela is a Rust-based malware analysis toolkit I’ve been building for a while — tools like tiqueryfileanalyzermstrings, and others. The MCP server exposes all of those tools to Claude Desktop natively, so instead of dropping to the terminal for every command, I can run analysis steps conversationally and let Claude help interpret the results and suggest next moves.

This is not replacing the terminal — it’s augmenting it. The pivot decisions still come from the analyst. But having a reasoning layer that can look at mstrings output and say “that SetDllDirectoryW + GetTempPathW combination is staging behavior, and here’s the ATT&CK mapping” is genuinely useful when you’re moving fast.

Both samples were sitting in a folder on my Desktop. I had SHA-256 hashes. Let’s go.


Phase 1: Threat Intelligence Query

First move is always TI. The MalChela tiquery tool hits MalwareBazaar, VirusTotal, Hybrid Analysis, MetaDefender, and Triage simultaneously and returns a combined results matrix. Two calls, two answers.

Sample 1 (4f3b8971...) came back confirmed LunaStealer across all five sources. First seen 2025-12-01. Original filename sdas.exe. VT tagged it trojan.generickdq/python — already telling us something about the build.

Sample 2 (d4f57b42...) was more interesting. MalwareBazaar returned both LunaGrabber and LunaStealer tags. Triage clustered it with BlankGrabber, GlassWorm, IcedID, and Luca-Stealer. The original filename was loader.exe. That’s a different kind of name than sdas.exe. One sounds like a throwaway test artifact. The other sounds deliberate.

The TI results alone suggested these weren’t just two copies of the same thing. They were potentially different components of the same campaign.


Phase 2: Static PE Analysis

fileanalyzer and mstrings on both samples.

The first thing that jumped out was the imphash — f3c0dbc597607baa2ea891bc3a114b19 — identical on both. Same section layout, same section sizes, same import count (146), same 7 PE sections including the .fptable section that PyInstaller uses for its frozen module table. These two samples were compiled from the same PyInstaller loader template with different payloads bundled inside.

But the entropy diverged sharply. Sample 1 (sdas.exe) came in at 3.9 — low, even for a PyInstaller bundle. Sample 2 (loader.exe) was 6.9 — high, indicating the embedded payload is compressed or encrypted more aggressively. Combined with the file size difference (47 MB vs 22 MB), this was the first signal that what was inside each bundle was meaningfully different.

mstrings gave us 22–23 ATT&CK-mapped detections across both samples — largely the same set: IsDebuggerPresentQueryPerformanceCounterSetDllDirectoryWGetTempPathWExpandEnvironmentStringsWOpenProcessToken. Standard infostealer staging behavior. Tcl_CreateThread showed up in both, which is a PyInstaller artifact from bundling Python with Tkinter. The VT python family tag made more sense in context.


Phase 3: PyInstaller Extraction

Both samples were extracted with pyinstxtractor-ng. This is where the two samples started to diverge clearly.

Sample 1 entry point: sdas.pyc — Python 3.13, 112 files in the CArchive, 752 modules in the PYZ archive.

Sample 2 entry point: cleaner.pyc — Python 3.11, 113 files, 760 modules.

The name cleaner.pyc inside a file called loader.exe is a tell. That’s not a stealer payload name. That’s something that runs after.

The bundled library sets were nearly identical between both — requestsrequests_toolbeltCryptodomecryptographypsutilPILsqlite3win32 — same stealer framework. But Sample 2 had a unique addition: a l.js reference (mapped to T1059 — Command and Scripting Interpreter). A JavaScript component not present in the December build. The OpenSSL versions also differed: Sample 1 bundled libcrypto-3.dll (OpenSSL 3.x), Sample 2 had libcrypto-1_1.dll (OpenSSL 1.1). Different build environments, roughly one month apart.

At this point the working theory was solid: Sample 1 is a standalone stealer. Sample 2 is a later-generation dropper/installer with an updated payload and additional capability.


Phase 4: Bytecode Decompilation

decompile3 couldn’t handle Python 3.11 or 3.13 bytecode. That’s a known limitation. pycdc (Decompyle++) handles both.

sdas.pyc decompiled cleanly — the import stack made the capability set immediately obvious:

from win32crypt import CryptUnprotectData
from Cryptodome.Cipher import AES
from PIL import Image, ImageGrab
from requests_toolbelt.multipart.encoder import MultipartEncoder
import sqlite3

CryptUnprotectData for browser master key decryption. AES for the decryption itself. ImageGrab for screenshots. MultipartEncoder for structured exfiltration. Classic infostealer, nothing surprising.

cleaner.pyc was a different story. The decompiler output opened with this:

__________ = eval(getattr(__import__(bytes([98,97,115,101,54,52]).decode()), ...

Heavy obfuscation — byte arrays used to reconstruct evalgetattr, and __import__ at runtime so none of those strings appear in plain text. The approach is designed to evade static string detection. Decode the byte arrays and you get:

bytes([98,97,115,101,54,52]) → "base64"
bytes([90,88,90,104,98,65,61,61]) → b64decode("ZXZhbA==") → "eval"
bytes([90,50,86,48,...]) → "getattr"
bytes([88,49,57,112,...]) → "__import__"

Standard Python malware obfuscation. But buried further down in the decompile output was a large binary blob — a bytes literal starting with \xfd7zXZ. That’s the LZMA magic header.


Phase 5: LZMA Stage 2 Extraction

The blob was located at offset 0x17d4 in the pyc file. Extract and decompress it:

import lzma
blob = open('cleaner.pyc', 'rb').read()
idx = blob.find(b'\xfd7zXZ')
decompressed = lzma.decompress(blob[idx:])
# → 102,923 bytes

One important detail: the decompression is wrapped in a try/except LZMAError block with os._exit(0) on failure. If the decompression fails — as it would in some emulated sandbox environments — the process exits silently with no error. That’s the anti-sandbox mechanism.

The decompressed payload was another obfuscated Python source using a custom alphabet substitution encoding. The final execution chain was compile() + exec(). Decoding the full stage 2 revealed everything:

The injection URL:

https://raw.githubusercontent.com/Smug246/luna-injection/main/obfuscated-injection.js

This is the live Discord injection payload. The stage 2 pulls this JavaScript file from GitHub and injects it into the Discord desktop client’s core module, persisting across restarts.

The capability set from stage 2:

  • Anti-analysis checks on startup: process blacklist (~30 entries including wiresharkprocesshackervboxserviceollydbgx96dbgpestudio), MAC address blacklist (80+ VM prefixes), HWID blacklist, IP blacklist, username/PC name blacklists
  • Discord token theft from all three release channels (stable, canary, PTB)
  • Browser credential theft across 20+ Chromium and non-Chromium browsers
  • Roblox session cookie harvesting (.ROBLOSECURITY= targeting with API validation)
  • Desktop screenshot capture
  • Self-destruct: ping localhost -n 3 > NUL && del /F "{path}"

The ping delay is a simple trick — the 3-second wait lets the process fully exit before the delete fires, so the file removes itself cleanly after execution.


What MalChela + MCP Added to This Workflow

The honest answer is: speed and synthesis.

tiquery hitting five TI sources in one call versus five separate browser tabs or CLI invocations is a meaningful time saving, but that’s the surface benefit. The deeper value showed up in the mstrings step — getting ATT&CK-mapped output with technique IDs alongside the raw strings meant the behavioral picture came together faster than manually correlating imports against the ATT&CK matrix.

The MCP integration meant each of those steps — TI query, PE analysis, string extraction — could happen within the same conversation context. Claude could see the fileanalyzer output and the mstrings output together and note that the entropy difference between the two samples was significant, that the identical imphash meant shared loader infrastructure, that the staging imports in mstrings were consistent with the exfil approach suggested by the TI tags. That cross-tool synthesis is where the integration earns its keep.

The parts that still required manual work: pyinstxtractor-ngpycdc, the LZMA extraction, and decoding the stage 2. Those are terminal steps on the Mac.


IOCs at a Glance

Samples:

SHA-256FilenameFamily
4f3b8971...d0sdas.exeLunaStealer
d4f57b42...24loader.exeLunaGrabber

Injection URL:

https://raw.githubusercontent.com/Smug246/luna-injection/main/obfuscated-injection.js

Self-destruct pattern:

ping localhost -n 3 > NUL && del /F "{executable}"

Imphash (shared loader stub):

f3c0dbc597607baa2ea891bc3a114b19

A full IOC list including ~60 C2 IPs, MAC address blacklists, and HWID blacklists is in the analysis report linked below.

Downloads

  • 📄 [Full Analysis Report] — Complete investigation narrative, sample properties, capability breakdown, IOC documentation, campaign timeline, and recommendations. (LunaStealer_Analysis_Report.pdf)
  • 🛡️ [YARA Rules — PE] — Four rules targeting the PE samples: exact hash match, shared PyInstaller stub (imphash-based), infostealer payload strings, generic PyInstaller infostealer. (lunastealer_pe.yar)

If you’re running MalChela in your environment and want to reproduce the TI query steps, the MalChela MCP plugin source is on GitHub at github.com/dwmetz/MalChela. Questions or additions to the IOC list — find me on the usual channels.


MalChela Meets AI: Three Paths to Smarter Malware Analysis

In a previous post I wrote about integrating MalChela with OpenCode on REMnux and giving the AI a quick briefing on the tool suite so it could incorporate them into its analysis workflow. That was a promising proof of concept, but it raised a natural follow-up question: how do you make these integrations more robust, reproducible, and persistent?

Since that post, I’ve been experimenting with three different approaches to bringing MalChela into AI-assisted workflows — each suited to a different environment and use case. This post walks through all three.


Approach 1: The Kali MCP Server (Toby)

The first implementation started with Toby — my portable Raspberry Pi forensics toolkit running a customized Kali Linux build. Toby is designed for headless operation via SSH, which turns out to be exactly the right architecture for an MCP server. The developers of Kali recently added an update to support MCP integrations. (See https://www.kali.org/blog/kali-llm-claude-desktop/)

Model Context Protocol (MCP) is an open standard that allows AI assistants like Claude to interface with external tools and systems in a structured, reliable way. Instead of pasting instructions into a chat window each session, you define your tools once in a server configuration and the AI has consistent, persistent access to them.

The setup leverages an existing open-source mcp-kali-server that exposes Kali’s forensic and security tooling as MCP tools. On the client side (Mac), the claude_desktop_config.json simply points to Toby (or your Kali box) over SSH:

{
"mcpServers": {
"mcp-kali-server": {
"command": "ssh",
"args": [
"-i",
"/Users/dwmetz/.ssh/id_ed25519",
"dwmetz@192.168.10.89",
"mcp-server"
],
"transport": "stdio"
}
}
}

With this in place, Claude Desktop has persistent, session-independent access to Kali’s toolkit running on Toby. No need to re-brief the AI each session — the tools are always available and always described the same way.

Key prerequisite: passwordless SSH key-based auth between your Mac and Toby. If you haven’t set that up:

ssh-keygen -t ed25519
ssh-copy-id user@<toby/kali-ip>
# Then one manual SSH to accept the host key fingerprint
ssh user@<toby/kali-ip>

Adding MalChela to the Kali MCP Server

The mcp-kali-server ships with routes for Kali’s built-in security tools, but MalChela isn’t included out of the box. Adding it requires changes to two files: kali_server.py (the Flask API backend) and mcp_server.py (the FastMCP frontend). Both live at /usr/share/mcp-kali-server/.

How the architecture works: mcp_server.py is what Claude talks to — it defines MCP tool names, descriptions, and parameter schemas. When a tool is called, it POSTs to kali_server.py, which constructs the actual shell command and executes it on Toby. The critical detail is that MalChela’s binaries must be run from within the MalChela workspace directory — running them from an arbitrary working directory causes failures. The cd {MALCHELA_DIR} && prefix in every command handles this.

kali_server.py changes

Add the MALCHELA_DIR constant (update with your MalChela install path) and Flask routes after the existing tool routes, before the health check endpoint:

# ============================================================
# MalChela Tool Routes
# Adjust MALCHELA_DIR if MalChela is installed elsewhere
# ============================================================
MALCHELA_DIR = "/home/dwmetz/tools/MalChela"
@app.route("/api/tools/malchela/fileanalyzer", methods=["POST"])
def malchela_fileanalyzer():
"""MalChela: Hash, entropy, packing detection, PE headers, YARA scan, VirusTotal lookup."""
try:
params = request.json
filepath = params.get("filepath", "")
if not filepath:
return jsonify({"error": "filepath parameter is required"}), 400
command = f"cd {MALCHELA_DIR} && ./target/release/fileanalyzer \"{filepath}\""
result = execute_command(command)
return jsonify(result)
except Exception as e:
logger.error(f"Error in malchela_fileanalyzer endpoint: {str(e)}")
return jsonify({"error": f"Server error: {str(e)}"}), 500
@app.route("/api/tools/malchela/mstrings", methods=["POST"])
def malchela_mstrings():
"""MalChela: Extract strings, apply Sigma rules, map to MITRE ATT&CK."""
try:
params = request.json
filepath = params.get("filepath", "")
if not filepath:
return jsonify({"error": "filepath parameter is required"}), 400
command = f"cd {MALCHELA_DIR} && ./target/release/mstrings \"{filepath}\""
result = execute_command(command)
return jsonify(result)
except Exception as e:
logger.error(f"Error in malchela_mstrings endpoint: {str(e)}")
return jsonify({"error": f"Server error: {str(e)}"}), 500
@app.route("/api/tools/malchela/malhash", methods=["POST"])
def malchela_malhash():
"""MalChela: Query a hash against VirusTotal and MalwareBazaar."""
try:
params = request.json
hash_value = params.get("hash", "")
if not hash_value:
return jsonify({"error": "hash parameter is required"}), 400
command = f"cd {MALCHELA_DIR} && ./target/release/malhash \"{hash_value}\""
result = execute_command(command)
return jsonify(result)
except Exception as e:
logger.error(f"Error in malchela_malhash endpoint: {str(e)}")
return jsonify({"error": f"Server error: {str(e)}"}), 500
@app.route("/api/tools/malchela/nsrlquery", methods=["POST"])
def malchela_nsrlquery():
"""MalChela: Query file hash against NIST NSRL known-good database."""
try:
params = request.json
filepath = params.get("filepath", "")
if not filepath:
return jsonify({"error": "filepath parameter is required"}), 400
command = f"cd {MALCHELA_DIR} && ./target/release/nsrlquery \"{filepath}\""
result = execute_command(command)
return jsonify(result)
except Exception as e:
logger.error(f"Error in malchela_nsrlquery endpoint: {str(e)}")
return jsonify({"error": f"Server error: {str(e)}"}), 500
@app.route("/api/tools/malchela/hashit", methods=["POST"])
def malchela_hashit():
"""MalChela: Generate MD5, SHA1, and SHA256 hashes for a file."""
try:
params = request.json
filepath = params.get("filepath", "")
if not filepath:
return jsonify({"error": "filepath parameter is required"}), 400
command = f"cd {MALCHELA_DIR} && ./target/release/hashit \"{filepath}\""
result = execute_command(command)
return jsonify(result)
except Exception as e:
logger.error(f"Error in malchela_hashit endpoint: {str(e)}")
return jsonify({"error": f"Server error: {str(e)}"}), 500
@app.route("/api/tools/malchela/fileminer", methods=["POST"])
def malchela_fileminer():
"""MalChela: Scan a directory for file type mismatches and metadata anomalies."""
try:
params = request.json
dirpath = params.get("dirpath", "")
if not dirpath:
return jsonify({"error": "dirpath parameter is required"}), 400
command = f"cd {MALCHELA_DIR} && ./target/release/fileminer \"{dirpath}\""
result = execute_command(command)
return jsonify(result)
except Exception as e:
logger.error(f"Error in malchela_fileminer endpoint: {str(e)}")
return jsonify({"error": f"Server error: {str(e)}"}), 500

Important note on malhash: Unlike the other tools which take a file path, malhash takes a hash string as its argument. The route reads a hash parameter and passes it directly to the binary. Passing a filepath to malhash will fail silently — a subtle but critical distinction.

mcp_server.py changes

The MalChela tool definitions need to be added to the setup_mcp_server() function, immediately before the return mcp line.

    @mcp.tool(name="malchela_fileanalyzer")
    def malchela_fileanalyzer(filepath: str) -> Dict[str, Any]:
        """
        MalChela: Static file analysis - hashes, entropy, packing detection,
        PE metadata (imports, sections, timestamps), YARA matches, VirusTotal status.
        Best first step for any unknown file.

        Args:
            filepath: Absolute path to the file to analyze

        Returns:
            Analysis report
        """
        return kali_client.safe_post("api/tools/malchela/fileanalyzer", {"filepath": filepath})

    @mcp.tool(name="malchela_mstrings")
    def malchela_mstrings(filepath: str) -> Dict[str, Any]:
        """
        MalChela: String extraction with IOC detection and MITRE ATT&CK mapping.
        Applies Sigma-style detection rules, flags suspicious patterns (registry keys,
        encoded payloads, suspicious DLL+API combos), maps findings to ATT&CK techniques.

        Args:
            filepath: Absolute path to the file to analyze

        Returns:
            String analysis with ATT&CK mappings and IOCs
        """
        return kali_client.safe_post("api/tools/malchela/mstrings", {"filepath": filepath})

    @mcp.tool(name="malchela_malhash")
    def malchela_malhash(hash: str) -> Dict[str, Any]:
        """
        MalChela: Query a file hash against VirusTotal and MalwareBazaar.
        Returns detection ratio, AV verdicts, first/last seen dates, and sample metadata.
        Requires VT_API_KEY env var; MB_API_KEY optional.

        Args:
            hash: MD5, SHA1, or SHA256 hash string to query

        Returns:
            Threat intel results from VirusTotal and MalwareBazaar
        """
        return kali_client.safe_post("api/tools/malchela/malhash", {"hash": hash})

    @mcp.tool(name="malchela_hashit")
    def malchela_hashit(filepath: str) -> Dict[str, Any]:
        """
        MalChela: Generate MD5, SHA1, and SHA256 hashes for a single file.
        Lightweight when you need hashes without full fileanalyzer output.

        Args:
            filepath: Absolute path to the file

        Returns:
            File hashes
        """
        return kali_client.safe_post("api/tools/malchela/hashit", {"filepath": filepath})

    @mcp.tool(name="malchela_fileminer")
    def malchela_fileminer(folderpath: str) -> Dict[str, Any]:
        """
        MalChela: Scan a folder for file type mismatches and metadata anomalies.
        Classifies all files, flags executables disguised as other types.

        Args:
            folderpath: Absolute path to the folder to scan

        Returns:
            File classification and anomaly report
        """
        return kali_client.safe_post("api/tools/malchela/fileminer", {"dirpath": folderpath})

    return mcp

One parameter naming detail worth noting: mcp_server.py receives folderpath from Claude, but must pass it as dirpath in the JSON body to kali_server.py — that’s what the Flask route reads. The tool definition above handles this correctly.

After saving both files, restart the MCP server and restart Claude Desktop.

Testing on an unknown malware sample

[below is the exact response received from Claude]

I’ll run MalChela analysis on that file via Toby. Let me load the tools first.Tools loaded. Running fileanalyzer first for static analysis, then mstrings for IOC/ATT&CK mapping. Let me verify the path exists:File confirmed on Toby. The MalChela MCP tools use local paths — let me run it directly via the Kali execute_command since the MCP path routing doesn’t map Toby paths. I’ll invoke MalChela’s binaries directly:MalChela binaries found. Running the full suite now:That’s already very telling. Now running mstrings for IOC and ATT&CK mapping:Now let me pull the VirusTotal intel on the hash:That’s a solid picture. Here’s the full analysis breakdown:

🔴 MalChela Analysis — unknown.exe

Identification

FieldValue
SHA256dca40a0120ea176aa65b99c90d171761cb8bf696be4be026941f7f54e3c21635
MD55e364c1e35445b655ca30ab0c5bc7a4a
SHA10ac9950d00429ff4509c7fb9aa7320a45a0465ea
MalwareBazaar FilenameBL 3425005143.bat (disguised as .bat, submitted as .exe)
OriginPoland (PL), submitted anonymously
First Seen2026-03-03 13:38:42 UTC — very fresh

🧬 Verdict: GuLoader

39/69 detections on VirusTotal. Multiple vendors confirm this as GuLoader (NSIS-based downloader/injector):

  • ESET: NSIS/Injector.ENS trojan
  • Malwarebytes: Trojan.GuLoader
  • Fortinet: NSIS/Injector.7A4A!tr
  • Kaspersky: HEUR:Trojan-Downloader.Win32.Minix.gen
  • MalwareBazaar Signature: GuLoader

📦 Static Analysis Highlights

  • Entropy: 7.96 — near-maximum, strongly indicating packed/encrypted payload
  • Packer: NSIS (Nullsoft Scriptable Install System) wrapper — common GuLoader delivery vehicle
  • Compile Timestamp: 2018-12-15 (likely spoofed/stomped)
  • Unsigned — no digital signature
  • 5 PE sections including .ndata (atypical NSIS section, flagged as suspicious)
  • 159 imports, including the dangerous trio:
    • GetProcAddress / LoadLibraryExA — dynamic API resolution (evasion)
    • ShellExecuteExA — process execution

🎯 MITRE ATT&CK Mappings (from mstrings)

TechniqueIDTactic
Obfuscated Files or InformationT1027Defense Evasion
Process InjectionT1055Defense Evasion / Privilege Escalation
Application Layer Protocol: Web ProtocolsT1071.001Command and Control

🌐 Network IOC

  • http://nsis.sf.net/NSIS_Error — NSIS error handler URL (confirms NSIS wrapper), not inherently malicious but characteristic of GuLoader’s packaging

Summary

This is a live GuLoader sample — an NSIS-wrapped shellcode injector/downloader commonly used as a first-stage payload to retrieve second-stage malware (often AgentTesla, Remcos RAT, FormBook, etc.). The high entropy, dynamic API loading, and process injection indicators are all consistent with GuLoader’s known behavior. Treat this as active malware — do not execute outside an isolated sandbox.


Approach 2: Persistent MalChela Instructions in OpenCode on REMnux

The REMnux OpenCode integration I described in the previous post worked, but it had one friction point: each new session required re-pasting the MalChela context. OpenCode has a solution for this — the AGENTS.md file.

Place an AGENTS.md file in your working directory (or MalChela’s root directory) and OpenCode will automatically load it as persistent context for every session started from that location. This is the most lightweight way to make MalChela tool-aware without building any infrastructure.

Create the file at /home/remnux/Tools/MalChela/AGENTS.md:

# MalChela Tool Instructions for AI Assistants
MalChela is installed at: /home/remnux/Tools/MalChela
All tools are Rust binaries. The execution pattern is:
cd /home/remnux/Tools/MalChela
./target/release/<toolname> <arguments>
## Primary Static Analysis Tools
| Tool | Binary | Description |
|------|--------|-------------|
| File Analyzer | fileanalyzer | Hash, entropy, packing detection, PE info, YARA scan, VirusTotal lookup |
| mStrings | mstrings | String extraction, Sigma rule matching, Regex, MITRE ATT&CK mapping |
| NSRL Hash Lookup | nsrlquery | Query MD5/SHA1 against the NIST NSRL known-good database |
| Malware Hash Lookup | malhash | Query a hash against VirusTotal and MalwareBazaar |
## Additional Tools
| Tool | Binary | Description |
|------|--------|-------------|
| File Miner | fileminer | Scan directories for file type mismatches and metadata anomalies |
| Hash It | hashit | Generate MD5, SHA1, and SHA256 for a single file |
| mzHash | mzhash | Recursively hash all files in a directory |
| Extract Samples | extract_samples | Extract files from password-protected malware archives |
## Recommended Workflow
For initial triage of an unknown file:
1. fileanalyzer - establish baseline: hashes, entropy, PE headers
2. mstrings - extract strings, look for IOCs and ATT&CK technique indicators
3. malhash - check community threat intelligence
4. nsrlquery - confirm or rule out known-good status
## Environment Notes
- API keys (VT_API_KEY, MB_API_KEY) should be set in the shell environment
- MalChela integrates with REMnux tools; use REMnux CLI tools in conjunction as needed
- Case management is available via the MalChela GUI if a graphical session is active

When you start an OpenCode session from the MalChela directory, this context is automatically loaded. No manual pasting, no re-briefing.

Once complete I asked it to save the results to the Desktop in markdown.

What was cool to me about this approach is that using the installed REMnux tools, you can take the analysis further after MalChela. In my testing I (OpenCode) analyzed a file with mStrings. We then followed up by running capa against the file. From there we could compare what each tool detected and missed in the sample. (screenshot truncated)


Approach 3: MalChela as a Native MCP Server (Mac)

The most powerful integration is running MalChela as its own dedicated MCP server — making its tools directly available to Claude Desktop alongside the Kali server. This is what we built out as mcp-malchela.

The server is a small Node.js project that wraps MalChela’s binaries with proper MCP tool definitions. The key files are index.js (the server logic) and package.json.

package.json:

{
"name": "mcp-malchela",
"version": "1.0.0",
"description": "MCP server exposing MalChela malware analysis tools",
"main": "index.js",
"dependencies": {
"@modelcontextprotocol/sdk": "^1.0.0"
}
}

index.js — the server defines each MalChela tool with its input schema, executes the binary when called, and streams back results. Note that malhash is handled differently from the other tools — it receives a hash string rather than a filepath, so the argument routing accounts for that explicitly:

import { Server } from "@modelcontextprotocol/sdk/server/index.js";
import { StdioServerTransport } from "@modelcontextprotocol/sdk/server/stdio.js";
import { CallToolRequestSchema, ListToolsRequestSchema } from "@modelcontextprotocol/sdk/types.js";
import { execSync } from "child_process";
import path from "path";
const MALCHELA_DIR = process.env.MALCHELA_DIR || "/Users/dmetz/GitHub/MalChela";
const RELEASE_DIR = path.join(MALCHELA_DIR, "target", "release");
const server = new Server(
{ name: "mcp-malchela", version: "1.0.0" },
{ capabilities: { tools: {} } }
);
const tools = [
{
name: "malchela_fileanalyzer",
description: "Analyze a file: hashes, entropy, packing detection, PE headers, YARA scan, VirusTotal lookup",
inputSchema: {
type: "object",
properties: {
filepath: { type: "string", description: "Absolute path to the file to analyze" }
},
required: ["filepath"]
}
},
{
name: "malchela_mstrings",
description: "Extract strings from a file, apply Sigma rules and Regex patterns, map to MITRE ATT&CK",
inputSchema: {
type: "object",
properties: {
filepath: { type: "string", description: "Absolute path to the file to analyze" }
},
required: ["filepath"]
}
},
{
name: "malchela_malhash",
description: "Query a hash against VirusTotal and MalwareBazaar",
inputSchema: {
type: "object",
properties: {
hash: { type: "string", description: "MD5, SHA1, or SHA256 hash string to query" }
},
required: ["hash"]
}
},
{
name: "malchela_nsrlquery",
description: "Query a file hash against the NIST NSRL known-good database",
inputSchema: {
type: "object",
properties: {
filepath: { type: "string", description: "Absolute path to the file to query" }
},
required: ["filepath"]
}
}
];
server.setRequestHandler(ListToolsRequestSchema, async () => ({ tools }));
server.setRequestHandler(CallToolRequestSchema, async (request) => {
const { name, arguments: args } = request.params;
const env = { ...process.env };
const binaryMap = {
malchela_fileanalyzer: "fileanalyzer",
malchela_mstrings: "mstrings",
malchela_malhash: "malhash",
malchela_nsrlquery: "nsrlquery"
};
const binary = binaryMap[name];
if (!binary) throw new Error(`Unknown tool: ${name}`);
// malhash takes a hash string; all other tools take a filepath
const arg = name === "malchela_malhash" ? args.hash : args.filepath;
const cmd = `cd "${MALCHELA_DIR}" && ./target/release/${binary} "${arg}"`;
try {
const output = execSync(cmd, { env, encoding: "utf8", maxBuffer: 10 * 1024 * 1024 });
return { content: [{ type: "text", text: output }] };
} catch (error) {
return { content: [{ type: "text", text: error.stdout || error.message }] };
}
});
const transport = new StdioServerTransport();
await server.connect(transport);

Setup:

mkdir ~/mcp-malchela
cd ~/mcp-malchela
# create index.js and package.json as above
npm install

claude_desktop_config.json — with both servers configured:

{
"mcpServers": {
"mcp-kali-server": {
"command": "ssh",
"args": [
"-i",
"/Users/dwmetz/.ssh/id_ed25519",
"dwmetz@192.168.10.89",
"mcp-server"
],
"transport": "stdio"
},
"malchela": {
"command": "node",
"args": ["/Users/dmetz/mcp-malchela/index.js"],
"env": {
"MALCHELA_DIR": "/Users/dmetz/Tools/MalChela"
}
}
}
}

Restart Claude Desktop and both MCP servers will be active and available in every session.


I'll load the MalChela tools and run analysis on that file on your Mac.
Searched available tools
Starting with fileanalyzer for the full static analysis pass.
Fileanalyzer
VirusTotal flags it as malicious. Let me pull the full VT/MalwareBazaar intel and then run strings analysis.
Malhash
Confirmed SalatStealer. Let me run strings analysis to surface IOCs and ATT&CK mappings.
Ran 5 commands, used malchela integration
That's a solid picture. Here's the full breakdown:

🔴 unknown3.exe — SalatStealer (Confirmed Malicious)

Identity

FieldValue
SHA256e0c90f401ae5d23a551f2d75a826cec7f187a6b7f0da2d24565c880e027ae9a3
MD5cfcd64360a0b0c3185f9a48b1dbe1bdc
FamilySalatStealer (Trojan.PWS.Salat)
VT Detections33/68 — confirmed malicious
First Seen2026-03-03 15:37 UTC (very fresh)
OriginNL (Netherlands), reported by abuse.ch
LanguageGo (Golang) — evident from stdlib and module paths
Architecturei386 (32-bit PE)
PackerUPX (MalwareBazaar notes this is the UPX-decompressed form; original packed SHA256: 30a50cc0f7...)
Code SigningAbsent
Compile TimestampEpoch zero (1970-01-01) — deliberately zeroed/spoofed

MITRE ATT&CK Coverage (11 Techniques)

TechniqueIDWhat was found
Obfuscated Files / PackingT1027DOS stub + packer artifact
OS Credential Dumping (LSASS)T1003.001String matches
WMI Execution/DiscoveryT1047ffwmic reference
Process InjectionT1055Injection-capable imports
Scripting InterpreterT1059JSON/script interpreter refs
Web Protocol C2T1071.001WebSocket + HTTP/2 libs, hardcoded domains
System Info DiscoveryT1082System enumeration strings
File & Dir DiscoveryT1083File path enumeration
Ingress Tool TransferT1105Download capability
Unsecured CredentialsT1552passwordusernameloginPBEpasswordSet

Notable IOCs from Strings

  • Clients\DiscordTokens.txt — Discord token harvesting
  • moz_cookies SQL query — Firefox cookie theft
  • yandexbrowser — Yandex Browser credential targeting
  • *main.loginPBE — Login/credential parsing function
  • WebSocket C2 comms — *websocket.netDialerFunc*websocket.netError
  • WMI usage — likely for system enumeration pre-exfil

Bottom Line

This is a Go-compiled password/credential stealer targeting browsers (Firefox cookies, Discord tokens, Yandex Browser), using WebSocket for C2, with WMI-based discovery. It was UPX-packed (this is the unpacked copy), has a zeroed compile timestamp to defeat timestamp-based triage, and was submitted to MalwareBazaar just hours ago — very fresh sample. Do not execute in any environment without proper isolation.

VT permalink for full AV breakdown: https://www.virustotal.com/gui/file/e0c90f401ae5d23a551f2d75a826cec7f187a6b7f0da2d24565c880e027ae9a3/detection


Comparing the Three Approaches

Kali MCP (Toby)OpenCode AGENTS.mdMalChela MCP
PlatformKali (Raspberry Pi)REMnuxmacOS
PersistenceFull (MCP server definition)Full (file-based)Full (MCP server definition)
Setup complexityMedium (SSH + server install + file edits)Low (one markdown file)Medium (Node.js server)
AI integration depthStructured tool callsNatural language with contextStructured tool calls
Best forField analysis with TobyREMnux-primary workflowsMac-based analysis with MalChela

All three approaches solve the same core problem from the original post: making the AI reliably aware of your tools without re-briefing it every session. The right choice depends on your primary analysis environment — or, if you’re like me, you end up running all three.


All of the configuration files discussed here — including the MCP server setup and the are available in the MalChela repository on GitHub. Clone or pull the latest and you should be ready to go. As always, if you run into issues or have ideas for where to take the MCP integration next, open an issue or drop me a note.