LightShip

io.github.lightshipHQ由 GitHub 帳號 lightshipHQ 發布v0.1.0-alpha.1更新於 Oct 24, 2026

Governed access to production AI-agent traces in an existing ClickHouse store.

發布方自己的服務:ship.page

已驗證Streamable HTTP可網頁執行資料庫安全與監控資料與分析

概覽

AI 產生的概覽

LightShip 讓助理在政策管控下存取存放在既有 ClickHouse 中的正式環境 AI 代理追蹤資料。

可以做什麼
  • 在 MCP 用戶端中針對正式環境代理的對話、工具呼叫與結果提出業務問題。
  • 依角色與屬性政策(編譯進 ClickHouse 查詢)調查真實追蹤,無需直接資料庫憑證。
  • 將大型調查的授權匯出寫入代理工作區,僅在對話中回傳路徑與數量。
  • 檢視針對既有追蹤表的排序鍵與索引建議(僅供參考)。
適用情境
適用於已將 OpenTelemetry 代理追蹤存入 ClickHouse 的團隊,希望讓領域專家或編碼代理在存取政策下檢視真實對話與工具呼叫,而不必發放資料庫憑證或把追蹤複製到另一個可觀測性平台。
執行需求
遠端 streamable HTTP 端點;驗證使用在 LightShip 中建立的 bearer API 金鑰,透過 Authorization 標頭傳送。管理員需先將 LightShip 綁定到既有 ClickHouse OpenTelemetry span 表並定義角色與政策。
安裝前請注意
實驗性 alpha,尚無穩定版本;應在隔離環境中以合成或非敏感資料評估。它不偵測 PII、不遮蔽內容,也不隱藏個別 span,且只要有一個符合的 span 即可存取整條追蹤。政策僅涵蓋經 LightShip 發出的請求,直接存取 ClickHouse 會繞過政策,查詢稽核為盡力而為。bearer API 金鑰屬於機密。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 LightShip,將其新增到工作區。
  2. 連接 Authorization token 憑證。它們會被私密儲存,絕不會在公開頁面上展示。
  3. 為需要使用其工具的對話啟用該服務。

Web executable,透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

交給代理程式安裝

把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式,或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南,先向你展示發布者、端點和登入方式,經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。

閱讀 https://sourceweft.com/mcp/io.github.lightshipHQ%2Flightship/install.md,並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前,先告訴我它的發布者、端點和登入方式,等我確認。不要把 API 金鑰寫進命令或我會提交的檔案;問我金鑰應從哪裡取得。

檢視代理程式所依據的安裝指南

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "lightship": {
      "type": "http",
      "url": "https://lightship-production.up.railway.app/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_TOKEN>"
      }
    }
  }
}

README

LightShip

Faster production-agent iteration, without a developer bottleneck.

[CI] [Release] [Container] [License]

LightShip gives domain experts governed, self-serve access to production agent traces. It sits in front of your existing OpenTelemetry data in ClickHouse, so humans and AI coding agents can investigate real conversations and tool calls without database credentials or another observability silo.

Documentation · Website · Contribute

Why LightShip

  • Find the patterns behind failures. Ask business questions across conversations, tool calls, and outcomes using Codex, Claude, or another MCP client.
  • Keep your existing trace store. Bind LightShip to your OpenTelemetry table without copying traces into a separate platform.
  • Instrument once; define policies later. Select existing columns and attribute keys as authorization inputs, then create or change policies without re-instrumenting. The same policy governs the web UI, REST API, and MCP.
  • Keep large investigations out of model context. The local MCP companion writes authorized exports into the agent workspace and returns only paths and counts to the conversation.

How it works

  1. An administrator binds LightShip to an existing OpenTelemetry span table in ClickHouse.
  2. They choose which columns and map keys may be used for search and access rules.
  3. LightShip compiles each caller’s roles and attributes into the ClickHouse query before returning any trace through the web UI, REST, or MCP.

A trace is the authorization unit: one matching span grants access to the complete trace. LightShip never gives restricted users direct ClickHouse credentials.

Performance

Query performance depends on the layout and indexes of your existing ClickHouse trace table. After you select the fields available to filters and access policies, LightShip reviews the table's sorting key and skip indexes and suggests DDL for missing indexes. These recommendations are advisory: LightShip never changes your ClickHouse schema automatically. See how to review the optimization report.

Get started

The documentation contains the maintained setup and reference material:

Need help deploying LightShip in your environment? Talk to us.

Security and project status

LightShip protects only requests sent through LightShip. Direct ClickHouse access bypasses its policies. It does not currently detect PII, redact payloads, or hide individual spans. Query auditing is best-effort, so an audit-write failure is logged without failing the trace request.

LightShip is an experimental alpha with no stable release. Evaluate it using synthetic or non-sensitive data in an isolated environment. Read the security guidance and current limitations before deployment.

Feedback

We want to help make autonomous work trustworthy and governable. If your team operates AI agents in production, we would love to hear what LightShip should support next.

Talk to us or open a GitHub issue.

Contributing

See CONTRIBUTING.md for development and testing, SECURITY.md for vulnerability reporting, and CHANGELOG.md for changes. LightShip is licensed under Apache-2.0.

來源:README.md,提交 0ff8db6

工具

0
SourceWeft 尚未讀取此伺服器的工具。

版本歷史

1
  1. v0.1.0-alpha.1最新Sep 16, 2026

評分

已滿足適用檢查項的 85%

該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通必需25 分
  • 已宣告認證方式必需10 分
  • 已列出工具必需15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分

其他 Ship 整合