DeepSeek Harness 插件

dsh-plugin-mcp-support

Thin, non-duplicating dsh wrapper over the native @deepseek-ai/dsh-mcp-client bridge: composition- and settings-driven MCP server mounts.(英文原文)

跳到安装方式

来源信息

GitHub 仓库
royenheart/dsh-plugin-mcp-support
最近更新
2026年8月16日
分类
工具与能力
GitHub stars
0
载体类型
plugin
目录证据
上游声明已找到 dsh.bundle
证据路径
package.json#dsh.bundle
核对版本
0.1.0-rc.8
上游核对日期
2026-08-20

该证据由上游目录提供。本站没有安装、运行或安全审核这个插件。

安装

默认先复制一段 Prompt,让 Agent 读 GitHub 仓库和源码;需要自己装时再切到命令。

复制这段 Prompt,发给 DSH、Codex 或其他 Agent,让它先读 GitHub 仓库和源码。

请先不要安装或执行任何命令。阅读这个插件的 GitHub 仓库、README 和关键源码,然后用清楚、直接的方式回答以下问题,帮助我判断它是否适合我的需求:

1. 这个插件是什么,解决什么问题;
2. 适合哪些用户和典型使用场景;
3. 安装后如何使用,并给出一个最小使用示例;
4. 有哪些已知限制,以及隐私、安全、兼容性或维护风险;
5. 给出“推荐 / 有条件推荐 / 不推荐”的明确建议和理由。

请区分仓库明确说明、根据源码推断和未知信息。证据不足时请明确说明,不要猜测或照抄 README。

GitHub:https://github.com/royenheart/dsh-plugin-mcp-support
插件名:dsh-plugin-mcp-support
作者:royenheart

检查来源文件

安装前先看这个插件目录里的 README 和其他文件。

文件资源管理器2 个文件
README.md来源说明 · 只读预览

@royenheart/dsh-plugin-mcp-support

A thin, non-duplicating wrapper over the native dsh MCP bridge @deepseek-ai/dsh-mcp-client. It makes MCP servers configurable through two layered sources:

1. Composition config — the plugin entry's servers list. 2. Persisted dsh settings — the mcp-support settings namespace.

The wrapper mounts one native mcp-client child fiber per effective server and re-syncs the mounted set whenever the settings section changes. It does not vendor or re-implement any connection, tool-discovery, or reconnect logic.

Layout

src/index.ts          # host plugin: settings registration + dynamic child mounts + status route
src/client/index.ts   # browser half: "mcp" view tab (status page)
src/core/config.ts    # pure normalize/merge helpers (no cordis imports)
src/core/status.ts    # pure status-row shaping helper
tests/                # node:test suite using a real cordis Context
lib/                  # built host + client entries (npm run build)

Web status view

The client half registers an mcp tab in the session header's view-tab row, immediately to the right of the 轨迹 (trajectory) tab and before any later tabs such as 技能. Selecting it fetches /plugins/@royenheart/dsh-plugin-mcp-support/status and shows each effective MCP server with its transport, mounted state, and the last mount error when present. No servers configured renders "No MCP servers configured."

Install into a profile

lib/ is generated locally and is not committed. install.py always builds the repository's own toolchain first (npm install when the toolchain is missing, then npm run build) and only reports an error when npm itself is missing.

Install/uninstall idempotently with the bundled script (stdlib-only Python). The package ships its own cordis.patch.yml (id mcp-support) and declares dsh.bundle.patch, so the script only links the package into the profile node_modules, adds the link: dependency, and appends the package to dsh.profile.bundles. The profile's own cordis.patch.yml is never modified:

python3 install.py install --profile web          # install
python3 install.py uninstall --profile web        # remove
python3 install.py install --profile web --home "$DSH_HOME"   # explicit home

Manual alternative:

dsh plugin --profile <profile-name> add link:/home/royenheart/projects/dsh-plugins/dsh-plugin-mcp-support

dsh plugin reconciles dsh.profile.bundles from the installed package's dsh.bundle declaration, so no profile patch edit is needed either.

Restart dsh. The plugin declares inject: ['settings', 'tools'], so it loads once both the dsh settings service and the native tool registry are available.

Composition config example

Configure the bundle-inserted row by id in the profile's own cordis.patch.yml (the bundle patch already inserts the row; this patch only overrides its config):

- id: mcp-support
  config:
    servers:
      - transport: stdio
        serverName: filesystem
        command: npx
        args:
          - -y
          - '@modelcontextprotocol/server-filesystem'
          - /tmp
        env: {}
        cwd: ''
        toolCallTimeoutMs: 60000
        failOnStartupError: false

      - transport: streamable-http
        serverName: everything
        url: http://localhost:3000/mcp
        headers:
          Authorization: Bearer secret
        toolCallTimeoutMs: 60000
        failOnStartupError: false

Settings example

Persisted settings are layered over the composition list. Servers are keyed by serverName: a settings server with the same name overrides the composition entry; settings-only servers are appended after composition servers.

The namespace is mcp-support. (If settings.yaml still carries the key from an earlier misspelled build, rename that key to mcp-support once.)

In the profile's settings document (settings.yaml):

mcp-support:
  servers:
    - transport: stdio
      serverName: filesystem
      command: npx
      args:
        - -y
        - '@modelcontextprotocol/server-filesystem'
        - /tmp

    - transport: streamable-http
      serverName: everything
      url: http://localhost:3000/mcp
      headers:
        Authorization: Bearer secret

Config reference

Each servers entry is exactly the native @deepseek-ai/dsh-mcp-client config union.

stdio

FieldRequiredDefaultNotes
transportyesstdio
serverNameyes[A-Za-z0-9_-]{1,32}, unique per source list
commandyesexecutable to spawn
argsno[]passed without shell interpolation
envno{}merged over the scrubbed ambient env
cwdno''child working directory
toolCallTimeoutMsno60000per-tool-call timeout
failOnStartupErrornofalsereject plugin activation on initial connection failure
reconnectnonative defaultsenabled, initialDelayMs, maxDelayMs, maxAttempts

streamable-http

FieldRequiredDefaultNotes
transportyesstreamable-http
serverNameyes[A-Za-z0-9_-]{1,32}, unique per source list
urlyesMCP endpoint URL
headersno{}extra request headers
toolCallTimeoutMsno60000per-tool-call timeout
failOnStartupErrornofalsereject plugin activation on initial connection failure
reconnectnonative defaultsenabled, initialDelayMs, maxDelayMs, maxAttempts

Develop

npm run typecheck
npm run build
npm test