BailingHub MCP Server
Enables submission, tracking, and retrieval of governed tasks from DeepSeek Harness against BailingHub routes, allowing job management without opening embedded chat interfaces.
README
BailingHub for DeepSeek Harness
简体中文 | English
Run governed tasks from DeepSeek Harness against a business route already connected to BailingHub, then track the same job and read its available result. You can do this from the Harness Web UI on your own computer without opening the embedded chat page in each business backend.
This is an independent community integration. It is not developed, certified, endorsed, or recommended by DeepSeek.
What You Get
After installation, DeepSeek Harness discovers three native tools:
| Harness tool | What it does |
|---|---|
mcp__bailinghub__submit_governed_job |
Submit task text with a stable request ID |
mcp__bailinghub__get_governed_job |
Read the current state and public result of one job |
mcp__bailinghub__wait_for_governed_job |
Wait briefly for the same job without resubmitting it |
BailingHub remains the control plane between the model and the business system. The URL, Client Token, and route are operator configuration, not model arguments. The model cannot switch to another route or supply administrator, executor, approval, or business-system credentials through these tools.
Install
Prerequisites:
- Node.js
22.19.0+or24+; pnpmand@deepseek-ai/dsh@0.1.0-rc.7;- a reachable BailingHub deployment;
- one BailingHub Client Token restricted to the route you want this Harness profile to use.
npm install --global pnpm @deepseek-ai/dsh@0.1.0-rc.7
dsh plugin --profile web add dsh-bailinghub@0.1.1
Configure the process that starts Harness:
export BAILINGHUB_BASE_URL='https://hub.example.com'
export BAILINGHUB_CLIENT_TOKEN='replace-with-a-route-scoped-client-token'
export BAILINGHUB_ROUTE='order_assistant'
Then verify the composed profile and start the local Web UI:
dsh --profile web --dump-config
dsh web
The default Web UI is local to your computer. Ask Harness to submit a task through
BailingHub, preserve the returned job_id, and wait for or query that same job.
For a first try, use a task supported by your configured route and say explicitly: submit it
once with a stable request_id, preserve the returned job_id, and query that same job if a
bounded wait times out.
Correct Task Flow
- Generate one stable
request_idfor one business request. - Call
mcp__bailinghub__submit_governed_jobonce. - Preserve its
job_id. - Call
wait_for_governed_job, or callget_governed_joblater. - A wait timeout is not a failed task. Do not submit a replacement request.
Identity and Permission Boundary
Using the local Harness UI removes the need to open a business backend's embedded chat UI;
it does not turn the Harness login or local user into a trusted business identity.
The current 0.1.x bundle deliberately does not accept an acting subject as tool input. A route or
downstream business system must still resolve trusted identity and perform final
authorization. If the selected action requires identity that the configured path cannot
establish, it should remain unavailable or be rejected.
This bundle governs only tasks submitted through its three BailingHub tools. It does not intercept or govern every other tool installed in DeepSeek Harness.
Supply-Chain Note
This bundle contains no custom runtime JavaScript, no production dependencies, and no
install scripts. When Harness starts the bundle, its built-in MCP Client runs the pinned
command npx -y --package=bailinghub-mcp-server@0.1.1 bailinghub-mcp-server outside the
agent sandbox. The first start may
need npm network access. Review and pin the package before using it in a sensitive
environment; production images may pre-cache the exact version.
Non-loopback HTTP is rejected by the MCP server by default. Only set
BAILINGHUB_ALLOW_INSECURE_HTTP=true on a controlled private network where TLS terminates at
another trusted boundary.
Compatibility and Feedback
The first release is verified against DeepSeek Harness 0.1.0-rc.7,
bailinghub-mcp-server@0.1.1, and BailingHub Client API v1. Harness is still a developer
preview, so each Harness release requires a compatibility smoke test.
Report problems through GitHub Issues. Never include tokens, private deployment URLs, personal information, or production business payloads.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。