---
title: "dsh-plugin-rea: REA for DeepSeek Harness"
url: "https://reamcp.dev/en/cases/dsh-plugin-rea-rea-for-deepseek-harness-c7a841"
source: "https://github.com/czcdr/dsh-plugin-rea"
source_type: "github"
repo: "https://github.com/czcdr/dsh-plugin-rea"
author: "czcdr"
author_handle: "czcdr"
stars: 3
category: "Tools & integrations"
categories: ["tools-integrations"]
tags: []
added: "2026-10-13T03:57:28.483Z"
updated: "2026-10-12T16:18:29Z"
locale: "en"
content_locale: "en"
---

# dsh-plugin-rea: REA for DeepSeek Harness

A third-party adapter that registers REA's MCP server inside DeepSeek Harness so its mcp__rea__* tools work there; analysis stays upstream.

## Description

Why it is listed: MIT-licensed integration glue.

## Original README

# dsh-plugin-rea

[![npm version](https://img.shields.io/npm/v/dsh-plugin-rea?style=flat-square&color=cb3837)](https://www.npmjs.com/package/dsh-plugin-rea)

把 **[REA — Reverse Engineer Anything](https://github.com/morluto/rea)** 接入 **DeepSeek Harness (dsh)**：
一个**纯配置型 Cordis bundle**，复用官方 MCP 客户端插件 `@deepseek-ai/dsh-mcp-client`，
把本机已安装的 `rea mcp`（stdio）注册成原生工具。装载后模型直接拥有 `mcp__rea__*` 工具。

> **与上游的关系（重要）**
> 本仓库是**第三方适配封装，不是 REA 项目的组成部分**，也不包含 REA 的任何源码，
> 更不由 REA 作者维护。它只做一件事：向 dsh 的配置组合里插入一条指向你本机 `rea` 安装的 MCP 服务器记录。
> - 上游项目：**REA — Reverse Engineer Anything** — <https://github.com/morluto/rea>（MIT）
> - 上游文档：<https://github.com/morluto/rea#what-you-can-analyze> · [中文](https://github.com/morluto/rea/blob/main/README_zh.md#可以分析什么) · 站点 <https://rea.tools/>
> - 上游 npm 包：`rea-agents`
> - 本插件仓库：**<https://github.com/czcdr/dsh-plugin-rea>** · npm：`dsh-plugin-rea`
> - 问题反馈：REA 本身的问题请提到上游；本插件的问题提到 [本仓库 Issues](https://github.com/czcdr/dsh-plugin-rea/issues)。

---

![dsh-plugin-rea：会话经官方 MCP 客户端行调用本机 rea mcp，再把 Evidence 带回会话](assets/screenshot-1.png)

## English

**What it does.** This plugin performs **no reverse engineering itself**. It connects the upstream REA
MCP server to DeepSeek Harness, so the tools REA ships become callable in a session as
`mcp__rea__*`.

In upstream's own words, REA is *"one MCP for reverse engineering across binaries, applications, and
runtime behavior"* — *"see a feature you like. Understand how it works, down to the binary level."* It
*connects your agent to tools for inspecting native binaries, JavaScript and Electron apps, .NET
assemblies, and websites*; analysis runs locally, and results include the evidence and limitations
behind each conclusion. Native analysis can use an existing Hopper, Ghidra, or IDA installation, and
static JavaScript analysis needs no native engine.

Upstream's own coverage list (`README.md#what-you-can-analyze`): native binaries — pseudocode,
assembly, strings, symbols, calls and references; JavaScript/Electron apps — modules, imports, source
maps, routes, IPC; websites — page structure, scripts, network observations, screenshots; .NET
assemblies; Android APKs; Android devices over adb; JEB projects; Android resources; firmware;
packages and assets; process behavior; plus offline ELF layout, EVM bytecode, recorded Linux crashes
and saved network captures.

**What this plugin adds.** Exactly two things: it registers that MCP server into DSH's tool layer
under the `mcp__rea__*` names, and it lets the engine environment (`GHIDRA_INSTALL_DIR`, …) be scoped
to that MCP child process instead of the whole harness.

**What it can and cannot do.** The analysis is upstream's; this section only restates upstream's own
contract so expectations are right.

- **It cannot recover complete source code.** For native binaries upstream returns *pseudocode and
  assembly*, for Android APKs *decompiled methods*, for JEB units *pseudocode*, and for
  JavaScript/Electron *modules and their relations*. Upstream's tool contract states the pseudocode
  "is decompiler output, not original source and not text-equivalent" to another provider's output.
- Not recoverable in practice: original identifier names, types, comments, macros and template
  expansion, build scripts and assets. Optimisation (inlining, reordering), symbol stripping,
  obfuscation, packing and VM protection only make the output worse.
- Every result carries **confidence and limitations** (`observed` / `derived` / `inferred`, plus an
  explicit limitations list). Read them as evidence, never as recovered source.
- Good for: understanding behaviour, locating an implementation, interoperability and compatible
  re-implementation, security review, migration reference. To obtain actual source code, use the
  legitimate route — the upstream repository, vendor authorisation, or the vendor's own SDK.

**Verified against** (author's machine, 2026-10-12/13):

| Component | Version |
|---|---|
| REA (`rea-agents`) | **6.4.0** |
| DeepSeek Harness (dsh desktop runtime) | **0.2.0-rc.2** |
| `@deepseek-ai/dsh-mcp-client` (row provider) | bundled by the harness; npm latest `0.0.1-rc.1` |
| Ghidra | 12.1.4 (REA currently accepts 12.1.x) |
| Node.js | 24.21.0 (REA engines: `^22.19.0 \|\| ^24.11.0 \|\| >=26.0.0`) |
| Host | Windows 11 x64 |

The shipped row was exercised live: `list_documents` answered through the `cmd.exe` form, and
`open_binary` completed with the `ghidra` provider while the engine directory came only from a
profile-level override.

**Prerequisites.** Node.js (a version REA accepts) and the REA CLI installed globally
(`npm i -g rea-agents`). For native analysis, Ghidra 12.1.x (Hopper / IDA are alternatives under
their own licences). Static JavaScript/Electron and .NET analysis need no engine.

**Install.**

```
dsh plugin --profile <profile> add github:czcdr/dsh-plugin-rea
# or, from npm:
dsh plugin --profile <profile> add dsh-plugin-rea
```

A prebuilt tarball is attached to [Releases](https://github.com/czcdr/dsh-plugin-rea/releases)
(version-free asset name, so the link keeps resolving).

**Configure — usually nothing.** The shipped row carries **no machine-specific path**:

- Windows: `command: cmd.exe`, `args: ['/c', 'rea', 'mcp']` — the npm shim is a `.cmd`, so a shell is
  required.
- macOS/Linux: `command: rea`, `args: ['mcp']` — see
  [`examples/cordis.patch.posix.yml`](examples/cordis.patch.posix.yml).

The only optional piece is the analysis engine. REA reads `GHIDRA_INSTALL_DIR` from its environment,
so either:

1. set it as an **OS-level environment variable** and (re)start the harness, or
2. pin it for the MCP child only — see
   [`examples/cordis.patch.with-env.yml`](examples/cordis.patch.with-env.yml).

**Keep machine-specific values out of the plugin.** Override the row from your own profile's
`cordis.patch.yml` (id-targeted config override). That file is machine-local and never published:

```yaml
- id: mcp-rea
  config:
    env:
      GHIDRA_INSTALL_DIR: '<your Ghidra 12.1.x directory>'
```

Only one row may use `serverName: rea` in a profile.

**Verify.** Ask for a `mcp__rea__*` call — `binary_session` reports the engine checks
(`configuration` / `installation` / `headless` / `native_decompiler` / `java`), all of which should be
`ok: true`. Then `open_binary` a PE, `binary_overview`, and `procedure_pseudo_code`.

**Scope and caveats.** REA exposes 170+ tools at 6.4.0, so every session's tool list grows; the row
sets `failOnStartupError: false` so a missing REA never blocks harness startup. Some REA tools launch
processes, write files and access the network — DSH's permission policy applies. Analyse only
artifacts you are authorized to inspect.

**Updating — two independent things, both manual.**

- **This plugin** never updates itself. Installed from a plugin market? Update it **in that market**.
  Installed from a local `file:` checkout? Overwrite the directory and install it again. Updating REA
  below does **not** update this plugin, and updating this plugin does not touch REA.
- **Upstream REA**: `npm i -g rea-agents@latest`. Its `bin` entry (`scripts/rea.mjs`) is stable across
  releases, so this needs no change here.
- **Ghidra** is version-gated by REA (12.1.x today): update REA first, then check that it accepts a
  newer Ghidra.
- **DeepSeek Harness**: a harness upgrade may change the MCP client's config schema — inspect the row
  with `Config.listConfigs name=@deepseek-ai/dsh-mcp-client` and adjust the field names.

`check-updates.ps1` (Windows PowerShell) reports the state of all four at once, and
`.github/workflows/validate.yml` enforces the bundle's shape — including "the shipped patch must not
contain a machine-specific path" — in CI.

**License.** MIT (this adapter). REA is MIT and owned by its authors; Ghidra is Apache-2.0 and owned
by the NSA. No upstream code is redistributed here.

---

## 中文

**作用。** 本插件**自身不做任何逆向分析**。它把上游 REA 的 MCP 服务接入 DeepSeek Harness，
让 REA 提供的那套工具在会话里变成可直接调用的 `mcp__rea__*`。

按上游自述，REA 是「**通过一个 MCP 服务，逆向分析二进制文件、应用程序和运行时行为**」——
「看到喜欢的功能，弄清它的工作原理，深入到二进制层面」。它「将智能体连接到分析工具，用于检查
原生二进制文件、JavaScript 和 Electron 应用、.NET 程序集以及网站」；分析在本机运行，结果包含
支撑各项结论的证据和相关限制。原生分析可以使用已有的 Hopper、Ghidra 或 IDA，静态 JavaScript 分析
不需要原生引擎。

上游列出的可分析目标（原文见 [README_zh.md#可以分析什么](https://github.com/morluto/rea/blob/main/README_zh.md#可以分析什么)）：
原生二进制文件——伪代码、汇编、字符串、符号、调用和引用；JavaScript / Electron——模块、导入、
source map、路由与 IPC；网站——页面结构、脚本、网络观察结果与截图；.NET 程序集；Android APK；
Android 设备（adb）；JEB 项目；Android 资源；固件；软件包与资源；进程行为；以及离线 ELF 布局、
EVM 字节码、已记录的 Linux 崩溃与已保存的网络捕获。

**本插件在此之上只增加两件事**：把该 MCP 服务注册进 DSH 的工具层（工具名 `mcp__rea__*`），
以及让引擎环境变量（`GHIDRA_INSTALL_DIR` 等）只作用于这个 MCP 子进程，而不是整个 harness。

**能做什么、不能做什么（重要）。** 分析能力由上游提供；这里只复述上游自己的契约，避免预期错位。

- **无法完整还原源代码。** 上游对原生二进制返回的是**伪代码与汇编**，对 Android APK 返回
  **反编译的方法**，对 JEB 单元返回**伪代码**，对 JavaScript/Electron 返回**模块及其关系**。
  上游的工具契约明确写着：伪代码是反编译器输出，**不是原始源码**，且与其他提供方的输出
  **不保证文本等价**。
- 实践中不可恢复的内容：原始标识符命名、类型、注释、宏与模板展开、构建脚本与资源；
  优化（内联、指令重排）、去符号（strip）、混淆、加壳与 VM 保护只会让可读性更差。
- 每条结论都自带**置信度与限制**（`observed` / `derived` / `inferred`，另有 limitations 列表）：
  请把它们当作证据使用，而不要当作"已还原的源码"。
- 适合做：理解行为、定位实现、互操作与兼容实现、安全审计、迁移参考。
  **要拿到真正的源码，请走正规渠道**——上游代码仓库、厂商授权，或厂商自己的 SDK。

**已验证版本**（作者机器，2026-10-12/13）：REA `6.4.0`、DeepSeek Harness 桌面运行时 `0.2.0-rc.2`、
Ghidra `12.1.4`、Node `24.21.0`、Windows 11 x64。行提供方 `@deepseek-ai/dsh-mcp-client` 由
harness 自带（npm 最新 `0.0.1-rc.1`）。发布用的这一行**实测跑通过**：`list_documents` 在
`cmd.exe` 形式下正常应答，`open_binary` 走 `ghidra` 提供方成功，而引擎目录仅来自本机 profile 覆盖。

**前置条件。** 满足 REA 要求的 Node；全局安装 REA（`npm i -g rea-agents`）。原生分析需要
Ghidra 12.1.x（Hopper / IDA 为替代方案，各自需授权）；静态 JavaScript/Electron 与 .NET 分析
不需要引擎。

**安装。**

```
dsh plugin --profile <profile> add github:czcdr/dsh-plugin-rea
# 或从 npm 安装：
dsh plugin --profile <profile> add dsh-plugin-rea
```

[Releases](https://github.com/czcdr/dsh-plugin-rea/releases) 附带预构建 tarball（资产名不带版本号，
链接不会失效）。

**配置——通常什么都不用改。** 发布出去的那一行**不含任何机器相关路径**：

- Windows：`command: cmd.exe`、`args: ['/c', 'rea', 'mcp']` —— npm 的 shim 是 `.cmd`，必须经 shell。
- macOS/Linux：`command: rea`、`args: ['mcp']` —— 见
  [`examples/cordis.patch.posix.yml`](examples/cordis.patch.posix.yml)。

唯一可选的是分析引擎。REA 从环境读取 `GHIDRA_INSTALL_DIR`，二选一：

1. 设为**系统级环境变量**，然后重启 harness（让 MCP 子进程继承）；
2. 只为这个 MCP 子进程指定 —— 见
   [`examples/cordis.patch.with-env.yml`](examples/cordis.patch.with-env.yml)。

**机器相关值不要写进插件。** 用你自己的 profile `cordis.patch.yml` **按 id 覆盖**这一行即可；
该文件只在你本机、永不随插件发布：

```yaml
- id: mcp-rea
  config:
    env:
      GHIDRA_INSTALL_DIR: '<你的 Ghidra 12.1.x 目录>'
```

同一 profile 内 `serverName: rea` 只能有一条记录。

**验证。** 调用任一 `mcp__rea__*` 工具：`binary_session` 会给出引擎检查项
（configuration / installation / headless / native_decompiler / java），应全部 `ok: true`；
随后 `open_binary` 打开一个 PE → `binary_overview` → `procedure_pseudo_code`。

**范围与注意。** REA 在 6.4.0 暴露 170+ 个工具，注入后每个会话的工具列表都会变大；行内设置
`failOnStartupError: false`，因此 REA 缺失也不会阻塞 harness 启动。部分 REA 工具会启动进程、
写文件、访问网络，按 DSH 的权限策略执行。请仅分析你已获得授权的目标。

**更新：两件事，且都要手动。**

- **本插件**不会自动更新。**从插件市场安装的，就在市场里更新**；本地 `file:` 目录方式安装的，
  覆盖目录后在插件管理器里重装。下面更新 REA **不会**更新本插件，反过来也一样。
- **上游 REA**：用 `npm i -g rea-agents@latest` 更新。它的 `bin` 入口（`scripts/rea.mjs`）
  跨版本稳定，所以本插件无需改动。
- **Ghidra**：受 REA 版本门槛限制（当前 12.1.x），先升 REA，再确认它接受新版 Ghidra。
- **DeepSeek Harness**：升级可能改变 MCP 客户端配置结构，用
  `Config.listConfigs name=@deepseek-ai/dsh-mcp-client` 看该行 schema 并调整字段名。

`check-updates.ps1`（Windows PowerShell）会把上述四项一次跑完；
`.github/workflows/validate.yml` 在 CI 里守住 bundle 的形状（含「发布物不得含机器路径」这一条）。

**许可。** 本适配器为 MIT；REA 为 MIT，版权归其作者；Ghidra 为 Apache-2.0，版权归 NSA。
本仓库不再分发上游代码。