re-llm-decompile
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., "@re-llm-decompiledecompile main function from the disassembly"
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.
re-llm-decompile
MCP server that exposes an AI decompiler via the OpenAI-compatible /v1/chat/completions API. Works with:
LLM4Decompile (the
EndorRefvariants, or the latest SK²Decompile) served via vLLMOllama running any code model (CodeLlama, DeepSeek-Coder, Qwen2.5-Coder, etc.)
Any other OpenAI-compatible endpoint
The server does not read files itself. The caller (Claude Code) is expected to obtain the disassembly with re-rizin.disassemble_function and pass it in. The server's job is to take the disassembly, optionally the raw bytes, and produce C-like pseudocode.
Tools
Tool | What it does |
| Hit |
| Send disassembly to the LLM, return C-like pseudocode |
| Have the LLM explain disassembly (no rewrite) |
| Have the LLM propose better names for compiler-generated symbols |
| Whole-binary summary from strings + imports + entry-point disasm |
Related MCP server: cutterMCP
Configuration
Env var | Default | Purpose |
|
| OpenAI-compatible base URL |
|
| Model name to request |
| (empty) | API key (use |
Install
This server is part of the RE-AI plugin. The plugin's install.sh / install.bat installs it as part of the standard flow.
To install standalone:
pip install -e ./servers/re-llm-decompileRun
re-llm-decompile # stdio transport (default for MCP)
python -m re_llm_decompile # equivalentUsage pattern (in Claude Code)
1. "Decompile main in /bin/ls"
2. Claude calls re-rizin.analyze_function → function list
3. Claude calls re-rizin.disassemble_function(name="main") → asm
4. Claude calls re-llm-decompile.decompile_function(asm=..., arch="x86_64")
5. Claude returns the C-like pseudocode to the userChoosing a model
LLM4Decompile 22B (Ref): best quality for Linux x86_64 binaries, requires ~44GB VRAM (or AWQ/GPTQ quantizations).
LLM4Decompile 6.7B (Ref): a good middle ground, ~14GB VRAM.
Ollama + Qwen2.5-Coder 7B: reasonable general-purpose code model. Quality is lower than LLM4Decompile for pure binary decompilation but it explains disassembly well.
Claude / GPT (via this server): not recommended — the prompt is tuned for open decompilation models. If you want to use Claude, call it directly through Claude Code rather than going through this server.
Deprecation note
The v1 re-ai repo did not have this server — it tried to decompile with pefile+capstone+llm prompts in its own agent loop. That is exactly the kind of thing Claude Code is good at. This server exists to give Claude Code a clean decompilation handle.
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.
Latest Blog Posts
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/Heretek-RE/re-llm-decompile'
If you have feedback or need assistance with the MCP directory API, please join our Discord server