ddsetup

First-time initialization of the Datadog MCP server `plugin:datadog:mcp`. When fulfilling requests that involve Datadog, use MCP tools from…

npx skills add https://github.com/datadog-labs/claude-code-plugin --skill ddsetup

Datadog MCP Server

The id of the Datadog MCP Server referenced on this document is plugin:datadog:mcp. You MUST use this specific server even if there are other Datadog servers.

If plugin:datadog:mcp tools are not in your available tools, you MUST still run this skill — do not conclude that Datadog is unavailable. Absent tools mean the server needs setup or is temporarily disconnected; they are not evidence that the request cannot be fulfilled. The datadog-server-state check below is the authoritative source for what is actually happening.

Accessing Datadog using other methods

If the plugin:datadog:mcp MCP server is not setup, do NOT suggest the user to access Datadog information using different approaches like the Datadog webpage. Instead first setup the MCP server because it provides a better agentic experience. Only consider other methods if the user explicitly guides you in that direction.

Shared reference

Read references/mcp-settings.md before proceeding. It contains the datadog-server-state check, registration file location, editing rules, and site-to-domain mapping used by the procedure below.

Setup procedure

Check the datadog-server-state (see mcp-settings.md):

  • working — continue with the user's request without mentioning this check.
  • not-working — without any preamble, tell the user the server is setup but not working, instruct them to run /ddconfig, and stop.
  • not-setup — the server needs first-time setup. Do not attempt to gather data using a different approach. Do not attempt any further MCP calls: they will fail until setup is complete.

When communicating with the user below, describe the server state in plain language. Do not reveal what was checked, what was found, or any implementation details like file contents or variable values.

What Datadog provides once set up

Datadog is an observability platform. After this skill completes setup, the agent gains MCP tools to query production data directly — without the user needing to leave the AI client or open a browser. Examples of what becomes possible:

  • Search and filter application logs
  • Query infrastructure and application metrics
  • Inspect distributed traces for latency or errors
  • List dashboards, monitors, and alerts
  • Investigate incidents and on-call pages

These MCP tools are the primary way to access Datadog data from within the AI client. Until setup is complete, none of these tools exist. The agent cannot see them, list them, or call them.

Steps

  1. Check for saved configuration. Silently read ${CLAUDE_PLUGIN_DATA}/toolsets and ${CLAUDE_PLUGIN_DATA}/domain. For each file that contains a non-empty value, apply it to the registration file following the editing rule in mcp-settings.md. Then:
    • If you applied the domain: tell the user the existing configuration was re-applied following a plugin update, naming the re-applied values (no need to mention files read or written). Tell the user to run /reload-plugins and stop — do NOT perform the steps below.
    • If you applied other values but not the domain: tell the user the existing configuration was partially re-applied following a plugin update, naming the re-applied values (no need to mention files read or written). Continue with the steps below.
    • If you applied nothing: continue with the steps below.

Now follow these steps to configure the domain:

  1. Ask for the domain. Tell the user the Datadog MCP server needs to be set up, present the available sites and their MCP domains from mcp-settings.md, and ask which domain to use. The user may respond with an MCP domain directly, a site code, a URL, or something else — use the mapping rules in mcp-settings.md to resolve the answer to an MCP domain. Ask for clarification if ambiguous.

    Follow the "Stay on script" rule in mcp-settings.md. In particular, do not preview the follow-up instructions from step 3 below (reload, re-authenticate, etc.) — that step emits them verbatim at the right moment.

  2. Apply the change. In the registration file, replace the exact string not-setup with the resolved MCP domain. Follow the editing rule in mcp-settings.md.

    Before:

    ${DD_MCP_DOMAIN:-not-setup}
    

    After (example for us1):

    ${DD_MCP_DOMAIN:-mcp.datadoghq.com}
    

    Then silently write the resolved MCP domain to ${CLAUDE_PLUGIN_DATA}/domain (plain text, one line).

  3. Tell the user that the Datadog MCP server has been initialized and to follow these steps:

    1. Run the command /reload-plugins
    2. Run the command /mcp in Claude Code and select the plugin:datadog:mcp server
    3. Select the authentication option

Thêm skills từ datadog-labs

dd-audit
datadog-labs
Điều tra dấu vết kiểm toán - ai đã thay đổi cái gì, xâm phạm khóa, nguyên nhân gốc rễ của sự tăng vọt chi phí, bằng chứng tuân thủ (SOC 2/PCI) và kiểm toán hoạt động AI.
official
agent-install
datadog-labs
Install the Datadog Agent on Kubernetes using the Datadog Operator — required before enabling Single Step Instrumentation (SSI), which automatically…
official
agent-observability-auto-experiment
datadog-labs
Chạy leo đồi cải thiện mã lặp đi lặp lại dựa trên dữ liệu Datadog LLM-Obs thực tế, cục bộ, với Claude Code làm tác nhân. Thiết lập đánh giá cơ sở, thực hiện một…
official
agent-observability-eval-bootstrap
datadog-labs
Khởi tạo bộ đánh giá từ dấu vết sản xuất — theo mặc định đề xuất bộ đánh giá LLM-judge trực tuyến và, sau khi bạn xác nhận, tạo chúng trong Datadog dưới dạng bản nháp bị vô hiệu hóa…
official
agent-observability-eval-pipeline
datadog-labs
Đường ống quan sát Agent từ đầu đến cuối cho ml_app được instrument hóa — phân loại các trace production, truy tìm nguyên nhân gốc rễ của sự cố, khởi tạo các bộ đánh giá, sau đó (tùy chọn)…
official
agent-observability-experiment-analyzer
datadog-labs
Phân tích kết quả thử nghiệm LLM. Xử lý các thử nghiệm đơn lẻ hoặc so sánh, chế độ khám phá hoặc hỏi đáp. Sử dụng khi người dùng nói "phân tích thử nghiệm", "so sánh…
official
agent-observability-replay-trace
datadog-labs
Sử dụng khi nhà phát triển muốn lặp lại trên MỘT trace Agent Observability / LLM Obs cụ thể mà họ không hài lòng với kết quả đầu ra — chạy lại trace đó với…
official
agent-observability-trace-rca
datadog-labs
Phân tích nguyên nhân gốc rễ trên các dấu vết LLM trong production. Chẩn đoán lý do ứng dụng LLM gặp lỗi — hoạt động từ phán quyết của bộ đánh giá, lỗi runtime, hoặc các vấn đề cấu trúc…
official