argus

argus

Reverse engineering against a running Windows process instead of the file on disk: memory reads, pattern and value scanning, disassembly of the code that actually executed, resolved IAT thunks, caller-chain tracing, and structure recovery. Complements the Ghidra and IDA servers rather than replacing them — correlate_addr maps a runtime address back to a module and RVA.

Category
Visit Server

README

Argus

English · 繁體中文 · 简体中文 · 日本語 · 한국어

An MCP server that gives AI agents evidence from a running process.

Every reverse-engineering MCP server so far bridges a static analyzer — Ghidra, IDA, apktool. They hand your agent the file on disk. None of them can tell the agent what the program is actually doing right now: what got decrypted into that buffer, which address the vtable slot resolved to at runtime, which call site sent that packet.

Argus is the other half. It attaches to a live Windows process and returns runtime evidence: real bytes at real addresses, disassembly of the code that actually executed, resolved IAT thunks, caller chains walked from live memory, and a hypothesis ledger that tracks what has been proven versus what is still a guess.

It does not draw conclusions for the model. It returns addresses, module/RVA context, instructions, callers, callees, and local evidence, then gets out of the way.


Why this exists

Games die. The publisher shuts the servers down, the studio folds, the genre moves on. What survives is a client sitting on somebody's hard drive that can no longer connect to anything — no source, no protocol documentation, no server left to talk to. Just an executable that still remembers how to speak a language nobody is listening to anymore.

Argus was built to bring those back.

Reconstructing a server for a dead game means recovering its protocol: packet layouts, encryption, opcode dispatch, the state machine on the other side of the wire. The only surviving specification is the client binary itself. Nobody wrote the document you need, and the people who knew have long since moved on.

Static analysis gets you partway. But a twenty-year-old client is packed, its strings are encrypted, its handlers dispatch through tables that only exist once the process is up. So you run it, and you watch it work — which is where the next section comes in.

That is what this tool is for: not breaking into something alive, but getting something dead to speak again.


Why go to the running process at all

A CPU cannot execute ciphertext.

Whatever a program does to protect itself on disk — packing, string encryption, virtualized instructions, imports resolved at load time — all of it has to be undone before the processor can run the code. At the instant of execution, the real instructions and the real data are sitting in memory in the clear. They have to be. That is not a flaw in any particular protector; it is a consequence of how processors work, and no amount of obfuscation gets around it.

So the two approaches are reading different things:

  • Static analysis reads the file. What the author shipped.
  • Runtime analysis reads what the file turned into. What the machine is actually running.

When those two differ, the second one is the truth.

Situation Static analyzer Argus
Packed / self-decrypting code sees the packer disassembles the unpacked bytes in memory
Indirect call through a vtable sees call [rax+0x18] resolves the slot to a concrete target
Import resolved at runtime sees a thunk stub resolves the thunk to the real API
Buffer contents after decryption nothing reads the plaintext
Which of 40 call sites actually fires guesses records the one that ran

Where static analysis wins

The trade runs the other way too, and it is worth being blunt about it: a static analyzer sees every path, including the ones that never execute. Argus only sees what actually ran. A branch that was never taken leaves no runtime evidence at all, and a function nobody called may as well not exist.

Neither view is complete on its own. That is what correlate_addr is for — map a runtime address back to a module and RVA, look it up in Ghidra or IDA, and work with both halves. Argus is built to sit alongside a static analyzer, not to replace one.


Tools

Process and memory processes_list · processes_find · mem_attach · mem_modules · memory_regions mem_read · mem_read_chain · mem_write

Scanning scan_bytes · scan_string · scan_regex · scan_pointers_to · scan_callers scan_x86_call_sites · value_scan_start · value_scan_refine · value_explain · real_rate

Disassembly and structure recovery disasm_at · analyze_function · find_vtable · extract_dispatch_tables analyze_send_call_sites · read_struct · diff_struct

Import and API resolution runtime_imports · runtime_exports · resolve_iat_thunks · resolve_api_targets

Tracing and correlation trace_call_chain · correlate_addr · locate

Evidence ledger record_hypothesis · verify_hypothesis · query_hypotheses · add_evidence


Two design decisions worth knowing about

Automatic architecture routing

Attaching a 64-bit analyzer to a 32-bit (WOW64) target is a classic source of silently wrong pointer arithmetic and garbage PE parsing. Argus ships a thin front-end, argus-router, which inspects the target process, determines whether it is x86 or x64, and dispatches to the matching argus-rs build. You configure one binary; the correct engine is selected per target.

The evidence ledger

Agents are good at producing plausible explanations and bad at noticing when a plausible explanation is unsupported. record_hypothesis / verify_hypothesis / add_evidence force the distinction: a claim is stored as a hypothesis, and only becomes an established fact when evidence is attached and verification passes. query_hypotheses lets a later session pick up where the previous one stopped without re-deriving everything.


Install

Prebuilt binaries

Download the latest release and unpack it anywhere:

Releases

The archive contains argus-router.exe plus both engine builds (argus-rs-x64.exe, argus-rs-x86.exe). Keep them in the same directory.

From source

Requires a Rust toolchain with both Windows targets installed:

rustup target add x86_64-pc-windows-msvc i686-pc-windows-msvc

git clone https://github.com/r0ptik/argus
cd argus
cargo build --release --target x86_64-pc-windows-msvc
cargo build --release --target i686-pc-windows-msvc

Configure

Claude Code

claude mcp add argus -- C:\path\to\argus-router.exe

Any MCP client

{
  "mcpServers": {
    "argus": {
      "command": "C:\\path\\to\\argus-router.exe"
    }
  }
}

Scope and intended use

Argus is a reverse-engineering and program-analysis tool. It is built for work such as game and server preservation, network protocol analysis, interoperability and clean-room reimplementation, malware analysis, crash and corruption debugging, and security research.

It is explicitly not built for attacking live services. No cheat features are accepted into this repository — see CONTRIBUTING.md.

It requires the ability to open and read another process, so use it only against processes you own or are authorized to analyze. Attaching to software you do not have permission to analyze may violate that software's terms or your local law. That is your responsibility, not the tool's.


Platform support

Windows only. The memory access layer (argus-winmem) is built on the Win32 process and memory APIs; there is no Linux or macOS backend today.

Both x86 and x64 targets are supported, including 32-bit processes running under WOW64.


Crates

Crate Role
argus-router front-end binary; architecture detection and dispatch
argus-rs MCP server; tool definitions and request handling
argus-engine analysis engine; disassembly, structure recovery, tracing
argus-winmem Win32 process and memory access
argus-scan pattern and value scanning primitives
evidence-core address, module, RVA and evidence data models

License

MIT. See LICENSE.

Recommended Servers

playwright-mcp

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.

Official
Featured
TypeScript
Magic Component Platform (MCP)

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.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

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.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

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.

Official
Featured
TypeScript
Kagi MCP Server

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.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

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.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured