ddtoolsets

作者: datadog-labs

Manages toolsets for the Datadog MCP server `plugin:datadog:mcp`. Use when the user wants to view, enable, or disable toolsets that control which tools are…

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

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.

Shared reference

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

Entry flow

Check the datadog-server-state (see mcp-settings.md). Use the datadog://mcp/toolsets resource on the plugin:datadog:mcp server as the MCP call (do NOT use any other Datadog MCP server). Do not output anything until the datadog-server-state and resource content are available, and proceed based on the results:

  • datadog-server-state=working AND valid content — without any preamble, go to the Toolsets Flow.
  • datadog-server-state=not-setup — without any preamble, tell the user the plugin is not set up and instruct them to run /ddsetup, and stop.
  • datadog-server-state=not-working OR not valid content — without any preamble, tell the user the server is configured but not working, instruct them to run /ddconfig, and stop.

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

Toolsets Flow

A toolset is a named group of related tools for a specific Datadog feature. Enabling a toolset makes its tools available; disabling it removes them.

How toolset defaults work

The DD_MCP_TOOLSETS default value in the registration file controls which toolsets are active. It has two states:

  • Empty (${DD_MCP_TOOLSETS:-}) — the server decides which toolsets to enable. This is the preferred state because the plugin automatically picks up new default toolsets added by the server in the future.
  • Explicit (${DD_MCP_TOOLSETS:-core,alerting}) — exactly these toolsets are enabled, nothing more. The server's defaults are ignored. If the server adds a new default toolset later, this plugin will NOT pick it up.

The order of toolsets in the comma-separated list is not meaningful. core,alerting and alerting,core are equivalent. When comparing lists (e.g. to check if the result matches the defaults), compare as sets, not strings.

When computing changes, always prefer empty over an explicit list that happens to match the current defaults. See the editing rule in mcp-settings.md for how to set an empty default value.

1. Gather toolset information

Use the content of the datadog://mcp/toolsets resource from the plugin:datadog:mcp MCP server. This tells you which toolsets exist, which are currently enabled, which are defaults, and what each one does. Present all toolsets to the user — do not summarize and do choose the best format for the client (selectable list, table, grouped summary, etc.). Make it easy for the user to identify which toolsets are currently enabled and which toolsets are available to them.

Also read the current DD_MCP_TOOLSETS default value from the registration file. If it is empty, the user is currently using server defaults. If it has an explicit list, those are the manually selected toolsets.

Any toolset name in the registration file that does not appear in the datadog://mcp/toolsets resource is unknown — ignore it when presenting to the user and silently drop it when writing the updated list.

2. Understand the user's intent

The user may want to:

  • Add more toolsets to the currently enabled list
  • Remove toolsets from the currently enabled list
  • Replace the entire list with a specific set of toolsets

Understand the user's intent from their response. Ask for clarification if ambiguous.

Important: If the current default value is empty (server defaults) and the user wants to add a toolset, you need to know what the defaults ARE so you can build the full list. Use the default information from the datadog://mcp/toolsets resource.

3. Compute the new toolset list

Apply the user's changes to produce a new comma-separated value for DD_MCP_TOOLSETS:

  • If the resulting list matches the default toolsets exactly → use an empty string (revert to server defaults).
  • If the user wants to revert to defaults (e.g. "reset", "use defaults") → use an empty string.
  • If all toolsets would be removed → use an empty string and warn the user that the server's default toolsets will be used instead.
  • If the resulting explicit list does not include core → warn the user before applying. The core toolset provides essential Datadog functionality and most workflows depend on it. Only proceed without core if the user explicitly confirms.
  • Otherwise → use the explicit comma-separated list.

4. Apply the change

Edit DD_MCP_TOOLSETS in the registration file following the editing rule in mcp-settings.md.

Example — adding alerting when currently using server defaults (assuming core and synthetics are defaults):

${DD_MCP_TOOLSETS:-}  →  ${DD_MCP_TOOLSETS:-core,synthetics,alerting}

Example — reverting to server defaults:

${DD_MCP_TOOLSETS:-core,alerting}  →  ${DD_MCP_TOOLSETS:-}

Then silently write the new DD_MCP_TOOLSETS value to ${CLAUDE_PLUGIN_DATA}/toolsets (plain text, one line — write an empty file if reverting to server defaults).

5. Confirm

Tell the user the toolsets have been updated including which toolsets are now enabled, and that they need 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

来自 datadog-labs 的更多技能

dd-audit
datadog-labs
审计追踪调查——谁更改了什么、密钥泄露、成本激增根本原因、合规证据(SOC 2/PCI)以及AI活动审计。
official
agent-install
datadog-labs
使用Datadog Operator在Kubernetes上安装Datadog Agent——这是在启用Single Step Instrumentation(SSI)之前所必需的,该功能会自动…
official
agent-observability-auto-experiment
datadog-labs
针对真实的Datadog LLM-Obs数据运行迭代式代码改进爬山算法,本地执行,以Claude Code作为代理。建立基线评估,做出一项…
official
agent-observability-eval-bootstrap
datadog-labs
从生产追踪中引导评估器——默认提议在线LLM-judge评估器,并在您确认后,在Datadog中将其创建为已禁用的草稿…
official
agent-observability-eval-pipeline
datadog-labs
End-to-end Agent Observability pipeline for an instrumented ml_app — classify production traces, root-cause failures, bootstrap evaluators, then (optionally)…
official
agent-observability-experiment-analyzer
datadog-labs
分析LLM实验结果。处理单一或对比实验,探索性或问答模式。当用户说“分析实验”、“比较…”时使用。
official
agent-observability-replay-trace
datadog-labs
当开发者想要针对某一条他们不满意输出的Agent Observability / LLM Obs trace进行迭代时使用——重新运行该trace以对比他们的……
official
agent-observability-trace-rca
datadog-labs
生产环境中LLM轨迹的根因分析。诊断LLM应用为何失败——基于评测裁判的判定、运行时错误或结构性……
official