Datadog MCP server: how founders with a coding agent connect it to Claude Code, Codex and Cursor
The Datadog MCP server is a remote server that lets a coding agent such as Claude Code, Codex or Cursor query your Datadog logs, traces, metrics, monitors and incidents. A founder connects it with an OAuth sign-in to the endpoint for their Datadog site. Datadog sets a fair-use burst limit of 50 tool calls per 10 seconds, with no monthly limit on calls, and lists no price for the server itself. It reads production telemetry and does not rank why visitors leave.
This page is for founders who build with Claude Code, Codex or Cursor and already send telemetry to Datadog, or are weighing it. We do not have a Datadog account. Everything about Datadog below is from Datadog's own pages, each marked with the date we read it. The setup steps are Datadog's, not a run of ours.
Jump to: setup · the tools · access and limits · price · where Datadog wins · where UserProof fits · sources · FAQ
How do you add Datadog's MCP server to a coding agent?
Datadog gives each site its own endpoint: add mcp. before the site name and /v1/mcp after it. For the US1 site (datadoghq.com) that is https://mcp.datadoghq.com/v1/mcp. EU1 is https://mcp.datadoghq.eu/v1/mcp. Datadog's docs list US3, US5, AP1, AP2 and UK1 the same way and mark the two US government sites, US1-FED and US2-FED, as not supported (setup page, read November 16, 2026).
| Agent | What Datadog's docs say to do | Then |
|---|---|---|
| Claude Code | Install Datadog's plugin: /plugin install datadog@claude-plugins-official. Without the plugin: claude mcp add --transport http datadog-mcp https://mcp.datadoghq.com/v1/mcp (US1; use your site's endpoint). | Run /ddsetup or ask any Datadog question, pick your Datadog site and finish OAuth. /ddtoolsets turns tool groups on and off. After a config change, run /reload-plugins. |
| Codex | Add to ~/.codex/config.toml: [mcp_servers.datadog] and url = "https://mcp.datadoghq.com/v1/mcp" (US1; use your site's endpoint). | Run codex mcp login datadog; it opens your browser for OAuth. To pick tool groups, add http_headers = { "X-Datadog-MCP-Toolsets" = "apm,llmobs" } on the next line. |
| Cursor | Install the Datadog Plugin from the Cursor Marketplace, or from Cursor Settings, Plugins, then Add to Cursor. | Type /ddsetup in the agent chat for first-time setup. Datadog's docs list no manual JSON snippet for Cursor on this page. |
The docs add that if you set up the server by hand before, you should remove that entry before installing a plugin, to avoid conflicts. For a custom Datadog sign-in subdomain, add ?subdomain=<SUBDOMAIN> to the endpoint. By default only the core toolset loads; add ?toolsets=... for more, or ?toolsets=all for every generally available toolset. Datadog notes that all uses more of the model's context window. If the connection shows anything other than connected, how to check if an MCP server is running reads each status line.
What can the Datadog MCP tools let an agent read?
Datadog's tools page groups tools into toolsets and says the tools are "under significant development and are subject to change" (tools page, read November 16, 2026). These are the ones that matter to a founder watching a site:
| Toolset | Tools | What the agent can read |
|---|---|---|
| core (default) | search_datadog_logs, analyze_datadog_logs, search_datadog_spans, aggregate_spans, get_datadog_trace, get_datadog_metric, search_datadog_monitors, search_datadog_incidents, search_datadog_hosts, search_datadog_dashboards | Logs, traces, metrics, monitors, incidents, hosts, dashboards, events and notebooks. |
| core and rum | search_datadog_rum_events, aggregate_rum_events | Real User Monitoring events: JavaScript errors, slow pages, counts and percentiles by page or country (Datadog's example prompts). |
| rum | search_rum_applications, get_rum_summary, get_rum_insight, get_rum_view_waterfall | Your RUM applications, vital metrics with period-over-period diffs, and view waterfalls. |
| session-replay | search_replays, get_replay_summary | Finds replays by user, device, error count or journey, then returns an AI-generated play-by-play of one session. |
| error-tracking | search_datadog_error_tracking_issues, get_datadog_error_tracking_issue | Error Tracking issues from RUM, logs and traces. |
| product-analytics | search_product_analytics_events and others | Product Analytics views and actions, aggregations, journeys, pathways and retention. |
| apm | apm_get_service_health, apm_investigate_errors, apm_latency_bottleneck_summary | Service health, errors and latency bottlenecks. Some APM tools are in Preview and need Preview access. |
| alerting (writes) | validate_datadog_monitor, create_datadog_monitor | Checks monitors and creates them in draft mode (Datadog says drafts send no notifications; you publish in the Datadog UI). create_datadog_monitor is the write tool. |
The tools page lists a session-replay toolset, but the setup page's list of toolsets does not include that name, so Datadog's docs don't say how to switch it on. The tools page names the permissions: search_replays needs RUM Apps Read, and get_replay_summary also needs RUM Session Replay Read. The docs also warn that tool results can contain text set by users or monitored systems, such as tags, and that your agent may read it as instructions, so review what your agent proposes.
Who can use it, and what are the limits?
- Sign-in: OAuth 2.0 is the recommended method. For a server or CI, where OAuth can't be completed, Datadog's docs prefer a Personal Access Token or Service Access Token sent as a bearer header, with an API key plus application key as the other option.
- Permissions: tools need
mcp_readto read andmcp_writeto change things, plus the normal Datadog permission for the data. Users with the Datadog Standard Role have both MCP permissions by default. Organization admins can switch MCP Access and MCP Write Access off for everyone. - Your data and the model: the docs say the MCP Server does not send your Datadog data to a third-party AI provider; your AI client and its model decide what goes to your AI provider.
- Rate limit: a fair-use burst limit of 50 tool calls per 10 seconds and no monthly limit on tool calls.
- Regions: not supported on the two US government sites. The ChatGPT app is in Preview and US1 only; that does not affect Claude Code, Codex or Cursor.
- Logging: Datadog records every tool call in its Audit Trail with the tool name, arguments, user and client.
All of this is from Datadog's MCP Server and setup pages, read November 16, 2026. Datadog's docs don't state a general-availability label for the server as a whole; they call individual toolsets generally available or Preview.
What does the Datadog MCP server cost?
Datadog's MCP docs and its pricing page give the MCP Server no price of its own, and they don't say whether it is billed separately. What the agent reads is billed under Datadog products. Datadog's pricing page (pricing, read November 16, 2026) lists these for the replay side:
| Product | Starting at, billed annually | On-demand |
|---|---|---|
| RUM Measure | $0.15 per 1,000 sessions a month | $0.22 |
| RUM Investigate | $3 per 1,000 sessions a month | $4.50 |
| Session Replay (add-on to RUM Investigate) | $2.50 per 1,000 sessions a month | $3.60 |
Datadog's pricing FAQ says RUM Investigate keeps the sessions you retain for 30 days and that replay sessions are billed separately at $2.50 per 1,000 sessions. At 50,000 sessions a month with replays on, that is $125 for the replay add-on alone at the annual rate, before RUM Measure and RUM Investigate (arithmetic ours from Datadog's per-1,000 price). The whole bill, every SKU added up at 1,000 to 100,000 sessions, is on Datadog RUM and Session Replay pricing.
| Plan | Price | Recorded visits a month |
|---|---|---|
| Free | $0, 3 sites | 5,000 |
| Pro | $29 a month | 50,000 |
| Scale | $99 a month | 250,000 |
The two are not like for like. Datadog sells an observability platform and UserProof sells one thing. Compare line items only if replays are the part you need.
Where does Datadog win?
Datadog does far more than UserProof. Its MCP server reaches APM traces, logs, infrastructure, monitors, incidents, Kubernetes, security signals, RUM, Error Tracking and Product Analytics, with your existing roles and access controls carried through. If your site runs on servers you need to keep alive, or your team already pays for Datadog, this is the one to connect, and no replay tool replaces it. Datadog's product page says to feed live logs, metrics and traces into coding agents like Claude Code, Cursor and Codex when you investigate production issues (product page, read November 16, 2026).
What does the Datadog MCP server not tell a founder fixing conversion?
Datadog's docs describe an agent reading telemetry, replay summaries and analytics events. They do not describe a ranked list of where real visitors leave your signup or checkout, tied to the file and line of your code to change. You would build that yourself from the replay and analytics tools, and the replay tools need RUM and Session Replay recording set up and paid for first.
UserProof does that other job and nothing else. It records real visitor sessions with a script tag, and its MCP server runs locally from a plugin. It hands your coding agent one fix at a time with the file and line, ranked by your signup or checkout goal, then checks whether the shipped fix worked. It is not an observability suite: it has no APM, logs, heatmaps, surveys or replay dashboard for analysts. Field values are never recorded. In Claude Code:
claude plugin marketplace add GetUserProof/userproof && claude plugin install userproof@userproof && claude "/userproof:setup"
It also works from Codex and Cursor, as the session replay MCP server page explains. See pricing for the plans above. For how replay tools with an MCP server compare, each endpoint checked, read which session replay tools have an MCP server. The same setup for a product analytics tool is in the Amplitude MCP server guide. Other servers worth wiring in are in the best MCP servers for coding agents.
Sources, read November 16, 2026
- Datadog, Datadog MCP Server (https://docs.datadoghq.com/mcp_server/): clients, data handling, permissions, rate limit, Audit Trail
- Datadog, Set Up the Datadog MCP Server (https://docs.datadoghq.com/mcp_server/setup/): endpoints, per-agent setup, toolsets, authentication, required permissions
- Datadog, Datadog MCP Server Tools (https://docs.datadoghq.com/mcp_server/tools/): tool names, toolsets, permissions
- Datadog, Pricing (https://www.datadoghq.com/pricing/): RUM Measure, RUM Investigate, Session Replay
- Datadog, MCP Server and Pup CLI (https://www.datadoghq.com/product/ai/mcp-server/): the stated use with coding agents
- UserProof plans and setup: getuserproof.com/pricing, set up UserProof in Claude Code
We have no Datadog account, so every step above is quoted from Datadog's setup docs rather than run here.
FAQ
Does Datadog have an MCP server?
Yes. Datadog runs a remote MCP server, one endpoint per Datadog site, such as https://mcp.datadoghq.com/v1/mcp for US1. Datadog's docs say it gives AI clients such as Cursor, OpenAI Codex and Claude Code structured access to your Datadog data. Sign-in is OAuth 2.0 by default. Datadog's docs say it is not supported on the US1-FED and US2-FED sites (docs read November 16, 2026).
How do I add the Datadog MCP server to Claude Code?
Datadog recommends its plugin: run /plugin install datadog@claude-plugins-official, then /ddsetup, pick your Datadog site and finish the OAuth login. Without the plugin, run claude mcp add --transport http datadog-mcp followed by the endpoint for your site, for example https://mcp.datadoghq.com/v1/mcp for US1. This comes from Datadog's setup page; we did not run it.
Is the Datadog MCP server free?
Datadog's docs and pricing page do not give the MCP server a price, and they do not say whether it is billed on its own. The docs state a fair-use limit of 50 tool calls per 10 seconds and no monthly limit on tool calls. The data it reads comes from products Datadog bills, such as RUM and Session Replay (pricing page, read November 16, 2026).
Can the Datadog MCP server show my coding agent session replays?
Datadog's tools page lists two Session Replay tools. search_replays finds recordings by user, device, error count or journey. get_replay_summary returns an AI-written play-by-play of one session. They need RUM and Session Replay recording set up in Datadog first, and the Session Replay add-on is priced per 1,000 sessions.
Can the Datadog MCP server change things in my account?
Yes, if you allow it. Datadog's docs say write tools such as create_datadog_monitor need the mcp_write permission and the alerting toolset, and an organization admin can turn off the MCP Server or only its write tools. A read-only user's call to a write tool is rejected.
What can't the Datadog MCP server tell me about conversion?
Datadog's docs describe reading telemetry, replays and analytics events. They do not describe a ranked list of the pages where visitors drop off tied to the file and line of your code to change. UserProof does that one job: it records real visitor sessions and hands your coding agent one fix at a time with the file and line.