re-dotnet
MCP server for .NET assembly analysis. Vendor-neutral wrapper around AsmResolver and ILSpy for parsing and decompiling .NET assemblies.
README
re-dotnet
MCP server for .NET assembly analysis. Vendor-neutral wrapper around AsmResolver (PE + .NET metadata) and ICSharpCode.Decompiler / ILSpy (C# decompiler).
Why
The other RE-AI MCP servers (re-lief, re-rizin, re-capa) handle native PE, ELF, MachO, DEX, ART, OAT — but none of them read the .NET CLI header or the ECMA-335 metadata tables. A MonoLauncher-style launcher that ships its payload as a managed .NET assembly is invisible to those tools; the import hash is empty, the strings table is uninformative, and rizin only sees native wrapper code.
re-dotnet fills that gap. It enumerates TypeDef / MethodDef / FieldDef, walks the #US string heap, and decompiles a single class or method to C# on demand.
Architecture
The MCP server itself is pure Python. The heavy lifting is done by a small .NET 10 CLI binary, re-dotnet-cli, built from src/re_dotnet/dotnet/Re.Dotnet.Cli/:
Claude Code (MCP stdio)
│
▼
re-dotnet server (Python, this directory)
│ subprocess.run(...)
▼
re-dotnet-cli (self-contained .NET 10 binary, single-file publish)
│
├─ AsmResolver 6.x (PE + .NET metadata, obfuscation-robust)
└─ ICSharpCode.Decompiler 9.x (C# decompiler)
The subprocess boundary is intentional:
- No
pythonnet.pythonnetis a fragile ABI bridge that breaks on every Python minor. Process isolation is robust. - No Mono. .NET 10 is the only supported runtime.
- Single-file publish. The CLI is a self-contained binary; no
dotnet --infodance, noDOTNET_ROOTenv vars, no GAC.
The Python wrapper locates the binary via this fallback chain (see runner.py):
$RE_DOTNET_CLI_PATH(escape hatch for tests)<server>/bin/re-dotnet-cli(install.sh default)- The dev build under
src/re_dotnet/dotnet/.../bin/ re-dotnet-clion$PATH
Tools
| Tool | What it does |
|---|---|
check_dotnet |
Health check — return .NET runtime + AsmResolver + Decompiler versions |
parse_assembly |
Enumerate TypeDef rows; return assembly header + one row per type |
decompile_type |
Decompile a single class to C# via ILSpy |
decompile_method |
Decompile a single method to C# via ILSpy |
list_strings |
Walk field-default constant strings (the #US heap subset) |
get_entry_point |
Return the managed entry point (<Module>::.cctor or Main) |
Install
./install.sh builds the .NET CLI via dotnet publish and copies the resulting binary to servers/re-dotnet/bin/.
To build standalone:
cd servers/re-dotnet/src/re_dotnet/dotnet/Re.Dotnet.Cli
dotnet publish -c Release -o ../../../../bin
To run:
re-dotnet # stdio transport (default for MCP)
python -m re_dotnet # equivalent
Requirements
- .NET 10 SDK or runtime (
dotnet --version≥ 10.0) - 2 GB disk for the .NET SDK + the self-contained binary
Deobfuscation notes
The ILSpy decompiler produces clean C# for non-obfuscated assemblies. For binaries that have been through commercial obfuscation (control-flow flattening, string encryption, JIT-hooking), the decompiler will surface Unable to decompile ... errors. The remediation path is to run the .NET deobfuscator first (the OSS de4dot shells out, GPL-3.0 — the RE-AI plugin keeps it process-isolated so the plugin's own distribution stays clean) and then re-run decompile_type / decompile_method on the cleaned output.
Pairing with re-leak-scan
A .NET-style launcher that calls out to telemetry endpoints (Sentry, Logstash, Confluence, Google Drive) will have those URLs as field-default constants. Run re-leak-scan (T2.2) on the assembly, or call list_strings here and pipe the output through a regex filter for the four endpoint families.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。