← MCP

Datadog MCP ServerDatadog MCP Server

Datadog

MCPMCP 可观测Observability DatadogDatadog 远程托管Remote-hosted 闭源Closed source toolsetstoolsets partial lifecycle: active 核验Checked 2026-10-23

Datadog 自营的官方 MCP server（mcp.datadoghq.com/v1/mcp）：agent 排障时直查生产日志、指标、trace、incident。三条硬口径：凭据透传——RBAC 原样生效，server 不越权； 工具按 toolsets 分组按需开；GovCloud 不兼容（官方明写）。诚实标注：无开源仓；另有一条独立的本地 Code Security MCP Server，别混。

Datadog's self-run official MCP server (mcp.datadoghq.com/v1/mcp): agents query live production logs, metrics, traces and incidents while debugging. Three hard facts: credentials are passed through — RBAC applies as-is, the server cannot exceed the user; tools are grouped into toolsets, enable on demand; GovCloud is not compatible (stated outright). Honest note: no open-source repo; a separate local Code Security MCP Server exists — don't conflate them.

适合与不适合When to use it, when not to

适合：编码/运维 agent 在 IDE 里对着真实生产数据排障、incident 响应取上下文。不适合：GovCloud；要全离线（工具在远端）；把写权限当默认（写动作逐调用检查）。

Fits: coding/ops agents debugging against live production data in the IDE, incident-response context. Not a fit: GovCloud (stated incompatible); fully offline (tools live remotely); assuming write access by default (write actions are checked per call).

中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-23

出品属地Origin: 国外出品foreign-made　Datadog, Inc.（美国）。Datadog, Inc. (US).
国内使用Direct access: 大陆可直连direct OK　docs.datadoghq.com 直连实测 200（2026-10-23）；mcp.datadoghq.com 端点直连 401（存在需鉴权）——大陆可达性未逐条核验。docs.datadoghq.com measured 200 direct (2026-10-23); the mcp.datadoghq.com endpoint measured 401 (exists, auth required) — mainland reachability not individually verified.
备案状态CAC filing: 未见于备案名单not in filing list：未见于网信办公示名单（2026-10-23 检索）；该产品未面向大陆公众提供服务，备案不适用。not in the CAC public registry (searched 2026-10-23); the product does not serve the mainland public, so filing does not apply.

备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.

固定坐标系Fixed coordinate system

11 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.

模型与开放条件Model & openness: 它自己不调模型——被 MCP 客户端调用的接口实现。它把 Datadog 可观测数据（logs/metrics/traces/dashboards/monitors/incidents/security signals 等）包装成工具给 agent 用。官方文档口径：多数工具直查 Datadog 后端不经模型；少数语义检索/自然语言转查询类工具会用 Datadog 的 AI 供应商托管的模型，可整组织关闭。
It calls no model itself — an interface implementation called by MCP clients. It wraps Datadog observability data (logs/metrics/traces/dashboards/monitors/incidents/security signals, etc.) as tools for agents. The official docs' position: most tools query the Datadog backend directly without a model; a few semantic-search / natural-language-to-query tools do use models hosted by Datadog's AI providers, and can be disabled org-wide.
运行位置Where it runs: 官方托管远程 server（mcp.datadoghq.com/v1/mcp——直抓 401=存在需鉴权）；另有本地二进制线（coterm.datadoghq.com 安装脚本装 datadog_mcp_cli，stdio 连远程工具）。官方文档明写 not GovCloud compatible；HIPAA-eligible。
An official hosted remote server (mcp.datadoghq.com/v1/mcp — direct fetch 401 = exists, auth required); there is also a local-binary line (the coterm.datadoghq.com install script installs datadog_mcp_cli, stdio connecting to remote tools). The official docs state plainly not GovCloud compatible; HIPAA-eligible.
本地文件Local files: 不碰本机文件系统——作用域是 Datadog 云端数据。例外：官方另有一条独立的 Code Security MCP Server（本地跑 SAST/SCA/IaC/secrets/SBOM 扫描），文档明说那是窄范围的另一条线，不是本条目的主体。
Never touches the local file system — the scope is Datadog cloud data. Exception: the official docs carry a separate Code Security MCP Server (running SAST/SCA/IaC/secrets/SBOM scans locally); the docs say explicitly it is a narrow, separate line, not the subject of this entry.
关机后的任务Tasks after shutdown: 服务在 Datadog 侧；会话请求-响应，无后台监听语义。官方声明会收集 MCP 使用信息（工具名/参数/错误/用户标识，不含 prompt 与模型推理），存 120 天。
The service lives on Datadog's side; session request-response, no background-listening semantics. The official docs state that MCP usage information is collected (tool names/parameters/errors/user identifiers, excluding prompts and model reasoning), stored for 120 days.
工具与扩展Tools & extensions: 按 toolsets 分组（URL 参数 ?toolsets= 选择，默认 core）：core（logs/metrics/traces/ dashboards/monitors/incidents/hosts/services/events/notebooks）、alerting、apm（Preview）、 cases、dbm、ddsql（Preview）、error-tracking、feature-flags、llmobs、product-analytics、 networks、onboarding、security、software-delivery、synthetics、workflows；toolsets=all 开全部 GA 组。
Grouped by toolsets (selected via the URL parameter ?toolsets=, default core): core (logs/metrics/traces/ dashboards/monitors/incidents/hosts/services/events/notebooks), alerting, apm (Preview), cases, dbm, ddsql (Preview), error-tracking, feature-flags, llmobs, product-analytics, networks, onboarding, security, software-delivery, synthetics, workflows; toolsets=all enables every GA group.
上下文与记忆Context & memory: 无 server 侧跨会话记忆——状态在 Datadog 平台本身；官方建议用 toolsets 裁剪工具省上下文窗口。
No server-side cross-session memory — state lives on the Datadog platform itself; the official docs recommend trimming tools via toolsets to save context window.
权限与限制Permissions & limits: 权限面是这条目的核心：官方明写 server 把已认证用户自己的凭据转发给 Datadog API—— RBAC/Data Access Control/日志限制查询原样生效，server 不能给用户越权。写动作要对应权限（如 monitors_write），每次工具调用都检查；只读用户调写工具会被拒。组织管理员可在 Organization Settings 管全局 MCP 访问与写能力，可开 IP allowlist。
The permission surface is the core of this entry: the official docs state plainly that the server forwards the authenticated user's own credentials to the Datadog API — RBAC / Data Access Control / log restrictions apply unchanged; the server cannot grant a user anything beyond their own access. Write actions require the corresponding permission (e.g. monitors_write), checked on every tool call; read-only users calling write tools get rejected. Org admins can manage global MCP access and write capability in Organization Settings, and can enable an IP allowlist.
适合什么任务Best for: 适合：编码/运维 agent 排障时取生产实时上下文（日志/指标/trace/incident）、在 IDE 里对着真实数据调试。不适合：GovCloud 环境（官方明写不兼容）；要本地全离线（远程线要连 Datadog；本地二进制也只是本地鉴权壳，工具仍在远端）。
Fits: coding/ops agents pulling live production context (logs/metrics/traces/incidents) while debugging, debugging against real data inside the IDE. Not a fit: GovCloud environments (officially stated incompatible); needing fully local offline (the remote line must reach Datadog; the local binary is only a local auth shell — tools still live remotely).
传输方式Transport: Streamable HTTP（远程 https://mcp.datadoghq.com/v1/mcp）为主；本地二进制 datadog_mcp_cli 走 stdio（鉴权在本地完成，工具仍在远端）。核验：官方文档 setup 页（2026-10-23 直抓）。
Streamable HTTP (remote https://mcp.datadoghq.com/v1/mcp) is primary; the local binary datadog_mcp_cli uses stdio (authentication happens locally, tools remain remote). Verified against: the official docs setup page (direct-fetched 2026-10-23).
认证机制Authentication: OAuth 2.0（推荐，客户端自动走流程；自定义子域要加 ?subdomain= 参数）；或 PAT/SAT Bearer 头；或 DD_API_KEY + DD_APPLICATION_KEY 头（官方建议用带 scope 的服务账号 key）。端点直抓 401=存在需鉴权。
OAuth 2.0 (recommended, clients run the flow automatically; custom subdomains need a ?subdomain= parameter); or a PAT/SAT Bearer header; or DD_API_KEY + DD_APPLICATION_KEY headers (the official docs recommend a scoped service-account key). The endpoint's direct-fetch 401 = exists, auth required.
权限范围Permission scope: 读面大（logs/metrics/traces/dashboards/monitors/incidents 等按 toolsets 分组）；写动作（建 monitor、改 case 等）要对应写权限、逐调用检查。官方口径：MCP 工具调用全部进 Audit Trail（工具名/参数/用户/客户端），另有 datadog.mcp.session.starts 与 datadog.mcp.tool.calls 两个标准指标。
A large read surface (logs/metrics/traces/dashboards/monitors/incidents etc., grouped by toolsets); write actions (creating monitors, modifying cases, etc.) require the corresponding write permission, checked per call. The official position: all MCP tool calls go into the Audit Trail (tool name/parameters/user/client), plus two standard metrics, datadog.mcp.session.starts and datadog.mcp.tool.calls.

头号误解Top misconception

把 Datadog MCP 当「能越权的旁路」——官方明写凭据透传，用户看不到的数据 server 也拿不到Treating the Datadog MCP as a privilege-exceeding side door — the README states credentials pass through; data the user cannot see, the server cannot fetch
把 Code Security MCP Server 与主体混为一谈——那是本地代码扫描的另一条窄线，官方文档专门区分Conflating the Code Security MCP Server with the main line — that is a separate narrow local code-scanning server, explicitly distinguished in the docs
全量 toolsets 一把开——官方建议按需裁剪，工具组太多吃上下文窗口Enabling all toolsets at once — the official docs recommend scoping tools; too many eat the context window
在 GovCloud 里规划接入——官方明写 not GovCloud compatiblePlanning access in GovCloud — the docs state outright: not GovCloud compatible

未知项清单Known unknowns

各 toolset 内工具的完整清单（官方文档列了组名与部分工具，未逐条清点）
速率限制的当前数值（官方文档给过口径且明写「可能调整」，按 D9 与快照纪律不记具体数字）
本地二进制线的 Windows 支持细节（官方给下载指引，未逐条核验）

The complete tool list within each toolset (the official docs list group names and some tools, not enumerated item by item)
Current rate-limit numbers (the official docs gave figures and state plainly they "may change"; per our price-recording policy and snapshot discipline no specific numbers are stored)
Windows support details of the local-binary line (the official docs give download guidance, not verified item by item)

证据来源Evidence sources

判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-23，内容更新日content updated on 2026-10-23。

类型Type	名称Name	链接Link
docs	Datadog · 官方 MCP Server 文档（定位、toolsets、权限与合规口径，2026-10-23 直抓核验）Datadog · official MCP Server docs (positioning, toolsets, permission and compliance positions, direct-fetched 2026-10-23)	https://docs.datadoghq.com/mcp_server/
docs	Datadog · 官方 setup 文档（端点、OAuth/PAT/SAT 鉴权、本地二进制、速率限制）Datadog · official setup docs (endpoint, OAuth/PAT/SAT auth, local binary, rate limits)	https://docs.datadoghq.com/mcp_server/setup/
changelog	Datadog · 官方发布博客（remote MCP server 上线与工具清单）Datadog · official launch blog (remote MCP server launch and tool inventory)	https://www.datadoghq.com/blog/datadog-remote-mcp-server/

实测记录Measured runs

本档未在真实客户端逐项跑通；工具清单与权限来自官方 README。

This site has not completed measured runs. The test protocol is at protocols/mcp/_protocol.md.