Skip to main content
Glama
IDA Pro MCP
by mrexodia

IDA Pro MCP

简单的MCP 服务器允许在 IDA Pro 中进行氛围逆转。

https://github.com/user-attachments/assets/6ebeaa92-a9db-43fa-b756-eececce2aca0

视频的二进制文件和提示可在mcp-reversing-dataset存储库中找到。

可用功能:

  • check_connection() :检查 IDA 插件是否正在运行。

  • get_metadata() :获取有关当前 IDB 的元数据。

  • get_function_by_name(name) :通过名称获取函数。

  • get_function_by_address(address) :通过地址获取函数。

  • get_current_address() :获取用户当前选择的地址。

  • get_current_function() :获取用户当前选择的函数。

  • convert_number(text, size) :将数字(十进制、十六进制)转换为不同的表示形式。

  • list_functions(offset, count) :列出数据库中的所有函数(分页)。

  • list_globals_filter(offset, count, filter) :列出数据库中匹配的全局变量(分页、过滤)。

  • list_globals(offset, count) :列出数据库中的所有全局变量(分页)。

  • list_strings_filter(offset, count, filter) :列出数据库中匹配的字符串(分页、过滤)。

  • list_strings(offset, count) :列出数据库中的所有字符串(分页)。

  • decompile_function(address) :反编译给定地址的函数。

  • disassemble_function(start_address) :获取函数的汇编代码(地址:指令;注释)。

  • get_xrefs_to(address) :获取对给定地址的所有交叉引用。

  • get_xrefs_to_field(struct_name, field_name) :获取对命名结构字段(成员)的所有交叉引用。

  • get_entry_points() :获取数据库中的所有入口点。

  • set_comment(address, comment) :为函数反汇编和伪代码中的给定地址设置注释。

  • rename_local_variable(function_address, old_name, new_name) :重命名函数中的局部变量。

  • rename_global_variable(old_name, new_name) :重命名全局变量。

  • set_global_variable_type(variable_name, new_type) :设置全局变量的类型。

  • rename_function(function_address, new_name) :重命名函数。

  • set_function_prototype(function_address, prototype) :设置函数的原型。

  • declare_c_type(c_declaration) :从 C 声明创建或更新本地类型。

  • set_local_variable_type(function_address, variable_name, new_type) :设置局部变量的类型。

不安全的功能(需要--unsafe标志):

  • dbg_get_call_stack() :获取当前调用堆栈。

  • dbg_get_registers() :获取所有寄存器及其值。此函数仅在调试时可用。

  • dbg_run_to(address) :运行调试器到指定的地址。

  • dbg_exit_process() :退出调试器。

  • dbg_continue_process() :继续调试器。

  • dbg_enable_breakpoint(address, enable) :启用或禁用指定地址的断点。

  • dbg_list_breakpoints() :列出程序中的所有断点。

  • dbg_set_breakpoint(address) :在指定地址设置断点。

  • dbg_delete_breakpoint(address) :删除指定地址处的断点。

  • dbg_start_process() :启动调试器。

先决条件

Related MCP server: idalib-mcp-headless

安装

安装(或升级)IDA Pro MCP 包:

pip install --upgrade git+https://github.com/mrexodia/ida-pro-mcp

配置 MCP 服务器并安装 IDA 插件:

ida-pro-mcp --install

重要提示:请确保完全重启 IDA/Visual Studio Code/Claude,以使安装生效。Claude 在后台运行,您需要通过托盘图标退出。

https://github.com/user-attachments/assets/65ed3373-a187-4dd5-a807-425dca1d8ee9

快速工程

LLM 容易出现幻觉,所以你的提示需要具体。对于逆向工程来说,整数和字节之间的转换尤其成问题。下面是一个最简单的示例提示,如果你使用其他提示取得了不错的效果,欢迎随时发起讨论或提交问题:

你的任务是使用 IDA Pro 分析一个 Crackme 文件。你可以使用 MCP 工具来检索信息。一般来说,使用以下策略:

  • 检查反编译并添加注释以说明您的发现

  • 将变量重命名为更合理的名称

  • 如有必要,更改变量和参数类型(尤其是指针和数组类型)

  • 更改函数名称,使其更具描述性

  • 如果需要更多详细信息,请拆解该函数并添加注释和您的发现

  • 切勿自行转换进制。如有需要,请使用 convert_number MCP 工具!

  • 不要尝试暴力破解,任何解决方案都只能通过反汇编和简单的 Python 脚本来获得。

  • 创建一份 report.md,其中包含你的发现和最后采取的步骤

  • 当你找到解决方案时,提示用户使用你找到的密码进行反馈

这个提示只是第一次实验,如果您找到了改进输出的方法,请分享!

提高法学硕士(LLM)准确性的技巧

大型语言模型 (LLM) 是强大的工具,但它们有时会难以处理复杂的数学计算,或出现“幻觉”(即虚构事实)。请务必告知 LLM 使用conver_number MCP,并且您可能还需要math-mcp进行某些操作。

另外需要注意的是,LLM 在处理混淆代码时效果不佳。在尝试使用 LLM 解决问题之前,请先检查一下二进制文件,并花些时间(自动)删除以下内容:

  • 字符串加密

  • 导入哈希

  • 控制流平坦化

  • 代码加密

  • 反反编译技巧

您还应该使用 Lumina 或 FLIRT 之类的工具尝试解析所有开源库代码和 C++ STL,这将进一步提高准确性。

SSE 传输和无头 MCP

您可以运行 SSE 服务器来连接到用户界面,如下所示:

uv run ida-pro-mcp --transport http://127.0.0.1:8744/sse

安装idalib后,您还可以运行无头 SSE 服务器:

uv run idalib-mcp --host 127.0.0.1 --port 8745 path/to/executable

注意: idalib功能由Willi Ballenthin贡献。

手动安装

注意:本节适用于需要详细安装说明的 LLM 和高级用户。

手动安装 MCP 服务器(Cline/Roo 代码)

要自行安装 MCP 服务器,请按照以下步骤操作:

  1. 全局安装uv :

    • Windows: pip install uv

    • Linux/Mac: curl -LsSf https://astral.sh/uv/install.sh | sh

  2. 克隆此存储库,例如C:\MCP\ida-pro-mcp 。

  3. 导航到 Cline/Roo Code MCP 服务器配置(见屏幕截图)。

  4. 单击*“已安装*”选项卡。

  5. 单击配置 MCP 服务器,将打开cline_mcp_settings.json 。

  6. 添加ida-pro-mcp服务器:

{
  "mcpServers": {
    "github.com/mrexodia/ida-pro-mcp": {
      "command": "uv",
      "args": [
        "--directory",
        "c:\\MCP\\ida-pro-mcp",
        "run",
        "server.py",
        "--install-plugin"
      ],
      "timeout": 1800,
      "disabled": false
    }
  }
}

要检查连接是否有效,您可以执行以下工具调用:

<use_mcp_tool>
<server_name>github.com/mrexodia/ida-pro-mcp</server_name>
<tool_name>check_connection</tool_name>
<arguments></arguments>
</use_mcp_tool>

IDA插件安装

MCP 服务器启动时,IDA Pro 插件将自动安装。如果您禁用了--install-plugin选项,请按照以下步骤操作:

  1. 将src/ida_pro_mcp/mcp-plugin.py复制(而不是移动)到您的插件文件夹(Windows 上为%appdata%\Hex-Rays\IDA Pro\plugins )中。

  2. 打开 IDB 并单击Edit -> Plugins -> MCP来启动服务器。

与其他 MCP 服务器的比较

目前市面上有一些 IDA Pro MCP 服务器,但我出于以下几个原因创建了自己的服务器:

  1. 安装应完全自动化。

  2. 其他插件的架构使得快速添加新功能变得困难(太多不必要的依赖样板)。

  3. 学习新技术很有趣!

如果你想查看它们,这里有一个列表(按照我发现它们的顺序):

请随意打开 PR 来在此处添加您的 IDA Pro MCP 服务器。

发展

添加新功能的过程非常简单流畅。您只需在mcp-plugin.py中添加一个新的@jsonrpc函数,即可在 MCP 服务器中使用,无需任何额外的样板代码!以下视频演示了我在不到 2 分钟的时间内(包括测试)添加get_metadata函数的过程:

https://github.com/user-attachments/assets/951de823-88ea-4235-adcb-9257e316ae64

要测试 MCP 服务器本身:

uv run mcp dev src/ida_pro_mcp/server.py

这将在http://localhost:5173打开一个 Web 界面并允许您与 MCP 工具进行交互以进行测试。

为了测试,我创建了一个指向 IDA 插件的符号链接,然后直接向http://localhost:13337/mcp发送一个 JSON-RPC 请求。 启用符号链接后,您可以运行以下命令:

uv run ida-pro-mcp --install

生成直接提交到main的变更日志:

git log --first-parent --no-merges 1.2.0..main "--pretty=- %s"

Available Tools

43 tools
check_connectionB

Check if the IDA plugin is running

Inputs: no named parameters.

convert_numberC

Convert a number (decimal, hexadecimal) to different representations

Inputs: size (required), text (required).

create_stack_frame_variableC

For a given function, create a stack variable at an offset and with a specific type

Inputs: function_address (required), offset (required), type_name (required), variable_name (required).

data_read_byteA
Read the 1 byte value at the specified address.

Only use this function if `get_global_variable_at` failed.

Inputs: address (required).

data_read_dwordA
Read the 4 byte value at the specified address as a DWORD.

Only use this function if `get_global_variable_at` failed.

Inputs: address (required).

data_read_qwordA
Read the 8 byte value at the specified address as a QWORD.

Only use this function if `get_global_variable_at` failed.

Inputs: address (required).

data_read_stringA
Read the string at the specified address.

Only use this function if `get_global_variable_at` failed.

Inputs: address (required).

data_read_wordA
Read the 2 byte value at the specified address as a WORD.

Only use this function if `get_global_variable_at` failed.

Inputs: address (required).

declare_c_typeC

Create or update a local type from a C declaration

Inputs: c_declaration (required).

decompile_functionC

Decompile a function at the given address

Inputs: address (required).

delete_stack_frame_variableC

Delete the named stack variable for a given function

Inputs: function_address (required), variable_name (required).

disassemble_functionC

Get assembly code for a function

Inputs: start_address (required).

get_calleesB

Get all the functions called (callees) by the function at function_address

Inputs: function_address (required).

get_callersC

Get all callers of the given address

Inputs: function_address (required).

get_current_addressB

Get the address currently selected by the user

Inputs: no named parameters.

get_current_functionB

Get the function currently selected by the user

Inputs: no named parameters.

get_defined_structuresC

Returns a list of all defined structures

Inputs: no named parameters.

get_entry_pointsB

Get all entry points in the database

Inputs: no named parameters.

get_function_by_addressC

Get a function by its address

Inputs: address (required).

get_function_by_nameC

Get a function by its name

Inputs: name (required).

get_global_variable_value_at_addressB
Read a global variable's value by its address (if known at compile-time)

Prefer this function over the `data_read_*` functions.

Inputs: ea (required).

get_global_variable_value_by_nameA
Read a global variable's value (if known at compile-time)

Prefer this function over the `data_read_*` functions.

Inputs: variable_name (required).

get_metadataB

Get metadata about the current IDB

Inputs: no named parameters.

get_stack_frame_variablesC

Retrieve the stack frame variables for a given function

Inputs: function_address (required).

get_xrefs_toC

Get all cross references to the given address

Inputs: address (required).

get_xrefs_to_fieldC

Get all cross references to a named struct field (member)

Inputs: field_name (required), struct_name (required).

list_functionsB

List all functions in the database (paginated)

Inputs: count (required), offset (required).

list_globalsB

List all globals in the database (paginated)

Inputs: count (required), offset (required).

list_globals_filterB

List matching globals in the database (paginated, filtered)

Inputs: count (required), filter (required), offset (required).

list_importsB

List all imported symbols with their name and module (paginated)

Inputs: count (required), offset (required).

list_local_typesB

List all Local types in the database

Inputs: no named parameters.

list_stringsC

List all strings in the database (paginated)

Inputs: count (required), offset (required).

list_strings_filterB

List matching strings in the database (paginated, filtered)

Inputs: count (required), filter (required), offset (required).

read_memory_bytesA
Read bytes at a given address.

Only use this function if `get_global_variable_at` and `get_global_variable_by_name`
both failed.

Inputs: memory_address (required), size (required).

rename_functionC

Rename a function

Inputs: function_address (required), new_name (required).

rename_global_variableC

Rename a global variable

Inputs: new_name (required), old_name (required).

rename_local_variableC

Rename a local variable in a function

Inputs: function_address (required), new_name (required), old_name (required).

rename_stack_frame_variableB

Change the name of a stack variable for an IDA function

Inputs: function_address (required), new_name (required), old_name (required).

set_commentB

Set a comment for a given address in the function disassembly and pseudocode

Inputs: address (required), comment (required).

set_function_prototypeC

Set a function's prototype

Inputs: function_address (required), prototype (required).

set_global_variable_typeC

Set a global variable's type

Inputs: new_type (required), variable_name (required).

set_local_variable_typeC

Set a local variable's type

Inputs: function_address (required), new_type (required), variable_name (required).

set_stack_frame_variable_typeC

For a given disassembled function, set the type of a stack variable

Inputs: function_address (required), type_name (required), variable_name (required).

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    C2.9/5.0

    Scored across 43 tools

    Disambiguation3/5

    The tool set has clear functional groupings, but there is significant overlap between data reading tools (data_read_byte, data_read_word, etc.) and memory/global variable tools, with explicit guidance to prefer certain tools over others indicating ambiguity. Tools like list_functions and get_function_by_address serve distinct but related purposes that could cause confusion in selection.

    Naming Consistency4/5

    Most tools follow a consistent verb_noun or verb_noun_preposition pattern (e.g., get_function_by_address, set_global_variable_type), with clear action-object naming. Minor deviations exist, such as convert_number (verb_noun) vs. declare_c_type (verb_noun), but overall the naming is predictable and readable across the set.

    Tool Count2/5

    With 43 tools, the count is excessive for typical MCP server purposes, leading to potential cognitive overload and reduced usability. While IDA Pro is a complex domain, the tool surface could be consolidated (e.g., merging data_read_* variants) to improve coherence without losing functionality.

    Completeness5/5

    The tool set comprehensively covers IDA Pro's core functionalities, including disassembly, decompilation, data reading, variable and type management, cross-referencing, and metadata retrieval. It supports full CRUD/lifecycle operations for functions, variables, and types, with no obvious gaps for agent-driven analysis workflows.

    Maintenance

    ActivityMaintained
    ResponsivenessOutdated data

    Related MCP Connectors

    Related MCP Servers