English | 中文 | 日本語 | 한국어 | Русский | العربية

Cheat Engine MCP Bridge

Let AI agents read, write, scan, and debug any process through Cheat Engine.

120 MCP tools for memory analysis, reverse engineering, debugging, and game modding -- all driven by natural language through any MCP-compatible AI agent (Claude, Cursor, Copilot, etc).

[!NOTE] Thanks everyone for the stars, much appreciated! <3

Quick Start

Prerequisites

• Windows (required -- uses Named Pipes)
• Cheat Engine 7.x installed
• Python 3.10+ with pip

Step 1: Install Python Dependencies

pip install mcp pywin32

Step 2: Load the Bridge in Cheat Engine

Option A — Manual (per session):

1. Open Cheat Engine
2. File > Execute Script > select lua/ce_mcp_bridge.lua > Execute

Option B — Auto-load (recommended): Copy lua/ce_mcp_bridge.lua into Cheat Engine's autorun directory so it loads every time CE starts:

C:\Program Files\Cheat Engine 7.x\autorun\ce_mcp_bridge.lua

After loading, confirm output: [MCP v11.4.0] Server started on CE_MCP_Bridge_v99, then attach to your target process as usual.

Step 3: Configure Your AI Client

Add the following to your MCP client config (replace C:/path/to/cheat-engine-mcp with the actual clone path):

{
  "mcpServers": {
    "cheatengine": {
      "command": "python",
      "args": ["-m", "ce_mcp"],
      "env": {
        "PYTHONPATH": "C:/path/to/cheat-engine-mcp"
      }
    }
  }
}

Config file locations:

• Cursor: ~/.cursor/mcp.json
• Antigravity: ~/.antigravity/mcp.json
• Claude Desktop: %APPDATA%\Claude\claude_desktop_config.json

Step 4: Verify

Ask your AI agent: "Ping the Cheat Engine bridge"

Expected: {"success": true, "version": "11.4.0", "message": "CE MCP Bridge Active"}

What Can It Do?

Full API reference: ce_mcp/ai_docs/command_reference.md

Architecture

flowchart LR
    AI["AI Agent\n(Claude / Cursor / Copilot)"]
    PY["Python MCP Server\npython -m ce_mcp"]
    PIPE["Named Pipe\nCE_MCP_Bridge_v99"]
    LUA["Lua Bridge\nce_mcp_bridge.lua"]
    TARGET["Target Process"]

    AI -->|"MCP (JSON-RPC / stdio)"| PY
    PY <-->|"Binary-framed JSON-RPC"| PIPE
    PIPE <-->|"Worker Thread"| LUA
    LUA -->|"CE API calls\n(main thread sync)"| TARGET

• Python side (ce_mcp/): Translates MCP tool calls into named-pipe commands. Stateless -- all state lives in CE.
• Lua side (lua/ce_mcp_bridge.lua): Runs inside CE. Worker thread handles pipe I/O; commands execute on the main thread via thread.synchronize so all CE APIs are safe to call.
• Protocol: 4-byte little-endian length header + UTF-8 JSON-RPC 2.0 body.

Example Workflows

Find and freeze a value:

You: "Scan for gold: 15000"              → AI calls scan_all("15000")
You: "Gold changed to 15100"             → AI calls next_scan("15100"), finds 3 addresses
You: "What writes to the first one?"     → AI calls find_what_writes(addr, timeout=5)
You: "Generate an AOB injection for it"  → AI calls generate_aob_injection(addr)

Reverse-engineer a structure:

You: "What's at [[game.exe+0x1234]+0x10]?"
AI: reads pointer chain, identifies RTTI: CPlayerInventory
AI: "0x00=vtable, 0x08=itemCount(int:24), 0x10=itemArray(ptr)..."

Critical Configuration

[!CAUTION] You MUST disable: Cheat Engine > Settings > Extra > "Query memory region routines"

Leaving it enabled causes CLOCK_WATCHDOG_TIMEOUT BSODs when DBVM or anti-cheat conflicts with memory region scanning.

Project Structure

ce_mcp/                      # Python MCP Server package
├── __main__.py              # Entry point (python -m ce_mcp)
├── server.py                # FastMCP instance
├── client.py                # Named-pipe bridge client
├── protocol.py              # Shared constants
├── _compat.py               # Windows stdio patch
├── tools/                   # 120 MCP tools
│   ├── _helpers.py          # Shared tool utilities
│   ├── process.py           # Process, modules, version
│   ├── memory.py            # Read / write / allocate
│   ├── scanning.py          # Value & pattern scanning
│   ├── analysis.py          # Disassembly, structures, symbols
│   ├── debugger.py          # Breakpoints, debugger, threads
│   ├── cheat_table.py       # Cheat table operations
│   ├── dbvm.py              # DBVM hypervisor (Ring -1)
│   └── scripting.py         # Lua eval, AA, injection gen
└── ai_docs/                 # AI knowledge base
    ├── command_reference.md # Full API reference (all 120 tools)
    ├── tool_guide.md        # AI agent usage guide
    └── ce_lua_api.md        # CE 7.6 Lua API reference

lua/ce_mcp_bridge.lua        # Cheat Engine Lua bridge

tests/                       # Test suite
├── test_bridge.py           # Low-level pipe protocol tests
├── test_lang_display.py     # Multi-language dialog tests
└── test_safe_tools.py       # Comprehensive safe-tools tests

docs/                        # Human-facing translations
├── README_zh.md             # 中文
├── README_ja.md             # 日本語
├── README_ko.md             # 한국어
├── README_ru.md             # Русский
└── README_ar.md             # العربية

requirements.txt             # pip dependencies

Acknowledgements

Core code derived from cheatengine-mcp-bridge by miscusi-peek.

Disclaimer

This project is for educational and research purposes only. It demonstrates the capabilities of the Model Context Protocol (MCP) for software analysis automation. Do not use it for malicious hacking, cheating in multiplayer games, or violating Terms of Service.