Amplitude MCP Server: Cursor IDE Setup (2026)
Connect Amplitude's official MCP server to Cursor IDE: OAuth setup, the streamable-http config, available analytics tools, and the active-development caveats to know before you rely on it.
How do you set up Amplitude MCP in Cursor? Add an amplitude entry under mcpServers in Cursor's MCP settings pointed at https://mcp.amplitude.com/mcp with "transport": "streamable-http", restart or reconnect, and authorize through the browser OAuth prompt on first use. No API key to generate or paste — Amplitude's server is remote and OAuth-secured, and any existing Amplitude customer can connect from Cursor, Claude Code, or another MCP-capable client the same way.
Quick Reference
| Maintainer | Amplitude (official) |
| Remote endpoint | https://mcp.amplitude.com/mcp |
| Transport | Streamable HTTP |
| Auth | OAuth — browser sign-in on first connection |
| Tool categories | Analytics queries, dashboards/notebooks, feature flags, session replay, AI agent analytics, feedback |
| Status as of 2026 | Actively developed — some tools and settings still changing |
What the Amplitude MCP Server Actually Does
Amplitude's MCP server gives an AI assistant read access to the product analytics data your team already has in Amplitude — charts, dashboards, experiments, feature flags, metrics, and session replay — without anyone exporting a CSV or screen-sharing a dashboard first. Once connected, asking Cursor something like "what's the week-over-week retention trend for the onboarding cohort" runs an actual query against your Amplitude project rather than guessing from whatever's in the chat history.
The tool surface is organized into a few groups, based on Amplitude's own documentation and server guide:
search, query_chart, query_experiment, query_dataset, query_metricget_charts, get_experimentsget_flags, create_flags, update_flagget_session_replays, list_session_replays, get_session_replay_eventsget_feedback_trendsThat's a broad surface for a server that's still actively developed — worth keeping in mind before you build a workflow that depends on a specific tool name staying stable release to release.
Prerequisites
url-based streamable-http servers natively)Step 1: Add the Server to Cursor
Open Cursor's MCP settings — Cursor Settings → Tools & Integrations → New MCP Server (or edit mcp.json directly) — and add:
{
"mcpServers": {
"amplitude": {
"url": "https://mcp.amplitude.com/mcp",
"transport": "streamable-http"
}
}
}
There's no header, token, or environment variable to fill in for the standard path. Everything after this is handled by the OAuth flow in Step 2.
Step 2: Authorize via OAuth
On the first tool call, Cursor opens a browser window prompting you to sign into Amplitude and approve the connection. Approve it, and Cursor stores the resulting credential for future sessions — you won't see the browser prompt again unless the authorization is revoked or expires.
Step 3: Verify the Connection
In Cursor chat, try something concrete rather than a vague "test the connection" prompt:
List the dashboards in my Amplitude project
or
Query the retention metric for the last 30 days and summarize the trend
A real answer referencing your actual project data means the connection and OAuth grant are both working. If Cursor says it has no Amplitude tools available, recheck the MCP settings panel for a green status indicator before assuming the query itself failed.
Practical Workflows
Debugging a metric regression:
Query the conversion funnel for the signup flow over the last 14 days
and tell me which step's drop-off rate changed the most.
Cross-checking a feature flag before shipping code:
Show me the current state of the "new-checkout" feature flag and
which segments it's currently enabled for before I merge this PR.
Pulling experiment results into a code review:
Get the results for the pricing-page experiment and summarize whether
the variant that this PR is about to ship met its success metric.
Reviewing session replay context for a bug report:
Find session replays from users who hit the checkout error in the
last 24 hours and summarize what they did right before the failure.
In each case, the value is the same: Cursor pulls real numbers from your actual Amplitude project instead of you tabbing over to the dashboard, screenshotting a chart, and pasting a description of it into chat.
The Caveat: This Is Still Actively Developed
Amplitude's own documentation describes the MCP server as under active development, with some functions and settings not yet available and the possibility of bugs or performance issues. That's a meaningfully different risk profile from a mature server like GitHub's or Stripe's official MCP integrations, which have had a year or more to settle. Treat any specific tool name or parameter in this guide as accurate as of this writing, but verify against Amplitude's own docs page if a call fails with an unfamiliar error — a tool getting renamed or restructured between releases is more likely here than on a longer-established server.
There's also a practical quota concern: Amplitude MCP calls run against the same API allowance your Amplitude plan already has. A broad, unscoped prompt like "pull every event for the last year and summarize trends" can burn through plan quota fast, the same way looping a script over the raw API would. Scope queries with a date range and a specific metric or chart rather than asking for everything and letting the model filter afterward.
Official Server vs. Community Wrappers
Amplitude's own mcp.amplitude.com endpoint, described above, is the one to start with — it's maintained by Amplitude, uses OAuth, and is documented on Amplitude's own site for Cursor specifically. You'll also find community-built Amplitude MCP wrappers on directories like LobeHub or Glama; most of these are read-only, wrap the same underlying Amplitude REST API with your own API key instead of OAuth, and don't add caching on top — meaning repeated similar queries cost the same API quota every time rather than being served from a cache. Unless you have a specific reason to run a community wrapper (a self-hosted requirement, for instance), the official OAuth-based server is the simpler and better-supported starting point.
Troubleshooting
OAuth popup never appears. Confirm Cursor is on a build that supports streamable-http remote servers, and that the transport field is set to "streamable-http" in the config — omitting it or misspelling it can cause the server to fail silently rather than surface a clear connection error.
Tool calls return "unauthorized" after previously working. OAuth grants can expire or be revoked from Amplitude's side. Disconnect and reconnect the server in Cursor's MCP settings to trigger a fresh authorization.
A tool that used to work now returns "tool not found." Given the server's active-development status, tool names and parameters can change between releases. Check Amplitude's MCP documentation for the current tool list rather than assuming your config is broken.
Queries feel slow or time out on large date ranges. Narrow the query — a specific metric or chart over two weeks resolves faster and more reliably than "everything from the last year." This also reduces the quota risk described above.
Multiple Amplitude projects, but Cursor only sees one. The OAuth grant is scoped to whichever project you authorized against in the browser flow. If your team has more than one Amplitude project, you may need a separate authorization (or a distinctly-named server entry) per project — check Amplitude's current documentation for whether project switching is supported within a single authorized session.
Frequently Asked Questions
Q: Do I need an API key for Amplitude MCP in Cursor?
A: No. The official server at mcp.amplitude.com uses OAuth — you authorize through a browser prompt once, and Cursor reuses that grant. Community-built wrappers that predate the official server typically use an API key instead, but the official OAuth path is the recommended starting point.
Q: What can Amplitude MCP actually do inside Cursor?
A: Query charts, metrics, datasets, and experiments; read and create dashboards and notebooks; read and update feature flags; pull session replay data; and, for teams instrumenting their own AI agents, query agent-specific analytics and feedback trends.
Q: Is the Amplitude MCP server stable enough for daily use?
A: It works for the documented use cases, but Amplitude's own docs describe it as under active development, meaning some functions may be incomplete and tool behavior can change between releases. Treat it as a productivity tool for exploration and debugging, not as an integration a critical pipeline depends on.
Q: Does using Amplitude MCP count against my Amplitude API quota?
A: Yes. MCP tool calls run against the same underlying Amplitude API your plan already has an allowance for. Broad, unscoped queries consume quota faster than filtered ones — always give the model a date range and a specific metric or chart rather than asking for everything.
Q: Can other MCP clients besides Cursor use this server?
A: Yes. Amplitude's server works from any MCP-capable client, including Claude Code, ChatGPT, Lovable, and Kiro, using the same OAuth-based connection — Cursor isn't special-cased.
Q: What's the difference between the official server and a community Amplitude MCP wrapper?
A: The official server is maintained by Amplitude, uses OAuth, and covers the tool categories described above. Community wrappers typically use your own Amplitude API key, are read-only, and don't cache results, so repeated queries cost the same quota each time. Start with the official server unless you have a specific reason to self-host.
Related Guides
Related guides
- Atlassian MCP Server Cursor IDE Setup 2026: Jira, Confluence & Bitbucket in One Config
- Attio MCP Server Cursor IDE Setup (2026): OAuth, No API Key, 48+ CRM Tools
- Auth0 MCP Server Setup for Cursor IDE (2026): Manage Tenants, Apps, and Actions from Chat
- AWS MCP Server Setup for Cursor IDE (2026): Query S3, Lambda & CloudWatch from Chat