Openlayer

com.openlayer由 openlayer.com 發布官方發布者v0.16.1更新於 Oct 27, 2026

Inspect AI projects, traces, tests, and governance rules in Openlayer, and debug failing evals.

已驗證Streamable HTTP可網頁執行開發者工具安全與監控AI 與機器學習

概覽

AI 產生的概覽

託管的 Openlayer MCP 伺服器，讓助理檢視 AI 專案、追蹤、測試與治理規則，並排查失敗的評估。

可以做什麼
  • 讀取所選工作區中的專案、資料來源、追蹤、測試、結果與治理規則
  • 依需求建立或更新這些專案、測試與治理規則
  • 透過 Openlayer 修復迴圈排查失敗的測試
  • 查詢 Openlayer 文件、管線與測試結果
適用情境
適合在 Openlayer 上建置或維護 LLM 與代理程式應用時使用，讓助理不必切換到網頁介面，就能檢視評估結果、追蹤正式環境行為或處理失敗的測試。
執行需求
遠端 streamable HTTP 端點 mcp.openlayer.com；需要 Openlayer 帳號，首次呼叫工具時以 OAuth 登入並選擇工作區。搭配的技能套件另需 OPENLAYER_API_KEY，自架部署則需將伺服器指向自己的位址。
安裝前請注意
助理可以建立、更新與刪除工作區物件，例如專案、測試與治理規則；README 表示寫入與刪除工具會要求確認，請留意這些確認提示。工具呼叫只送往 Openlayer 的 API，存取範圍限於登入時選擇的工作區。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Openlayer，將其新增到工作區。
  2. 連接 所需的 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/com.openlayer%2Fmcp/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。 客戶端會在首次連接時完成 OAuth 登入。

{
  "mcpServers": {
    "openlayer": {
      "type": "http",
      "url": "https://mcp.openlayer.com/mcp"
    }
  }
}

README

Openlayer Skills

Agent Skills that teach AI coding assistants (Claude Code, Cursor, GitHub Copilot, etc.) how to integrate apps with Openlayer — the AI evaluation and observability platform.

Coding agents produce significantly better results with the skill installed, because they are conditioned to follow Openlayer's current best practices instead of guessing from memory (outdated SDK calls, wrong env vars, invented test configs).

Skills

SkillDescription
openlayerMain skill. Instrument code with tracing (monitoring), set up offline openlayer.json + tests.json evals (development), create tests/guardrails, gate CI/CD, run the MCP fix-loop, and access data and docs.

Installation

Claude Code plugin

bash
claude plugin marketplace add openlayer-ai/openlayer-skillsclaude plugin install openlayer@openlayer

Cursor plugin

/add-plugin openlayer

GitHub Copilot

GitHub Copilot reads .github/plugin/plugin.json and the same skills and .mcp.json as Claude Code.

In VS Code, add the marketplace to your settings.json, then install Openlayer from the Extensions view (search @agentPlugins):

json
{  "chat.plugins.marketplaces": ["openlayer-ai/openlayer-skills"]}

In the Copilot CLI:

bash
copilot plugin marketplace add openlayer-ai/openlayer-skillscopilot plugin install openlayer@openlayer

Other agents

Codex, Gemini CLI, Windsurf, opencode, and dozens more install through the skills CLI:

bash
npx skills add openlayer-ai/openlayer-skills --skill "openlayer"

For claude.ai, Copilot's cloud coding agent and code review, a manual symlink, or anything else, see the installation guide — it is the source of truth and stays current as these tools change.

Prerequisites

An Openlayer account and an API key:

bash
export OPENLAYER_API_KEY=...# self-hosted / local only:export OPENLAYER_BASE_URL=https://api.openlayer.com/v1

Find your API key in the Openlayer app under Workspace settings → API keys (see https://docs.openlayer.com/workspace-and-projects/find-your-api-key).

Usage

Once installed, the agent uses the skill automatically when relevant — for example:

  • Adding Openlayer tracing/monitoring to an LLM or agent app
  • Setting up offline evals (openlayer.json + tests.json) and pushing commits
  • Creating tests from the catalog and configuring guardrails
  • Gating CI/CD on eval results
  • Debugging failing tests via the Openlayer MCP fix-loop
  • Querying projects, pipelines, test results, or looking up Openlayer docs

MCP server

The plugin also connects your agent to the hosted Openlayer MCP server at https://mcp.openlayer.com/mcp (US accounts on app.openlayer.com; this is the .mcp.json default). The first time a tool is used, you sign in to Openlayer with OAuth and pick a workspace. The agent can then read projects, data sources, traces, tests, results, and governance rules in that workspace, and create or update them when you ask. Write and delete tools ask for confirmation.

EU accounts (eu.openlayer.com) must use https://mcp-eu.openlayer.com/mcp. Each hosted MCP server only accepts tokens from its own region's API, so an EU account pointed at the US server fails OAuth. Outside the regional cloud hosts the default stays https://mcp.openlayer.com/mcp.

Tool calls go only to Openlayer's API. Nothing is sent anywhere else. See the privacy policy. Self hosted deployments should point the server at their own URL, as the MCP docs describe.

Feedback & Requests

Something not working as expected, or want a new use case covered? Open an issue at https://github.com/openlayer-ai/openlayer-skills/issues or email [email protected].

來源：README.md，提交 3ee3885

工具

0
該伺服器需要登入，因此 SourceWeft 無法在沒有帳號的情況下讀取其工具，也未收錄。已連接帳號所能看到的內容可能因帳號及其權限而異。

版本歷史

1
  1. v0.16.1最新Oct 20, 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