Amplitude MCP Server

Official Amplitude MCP for analytics, experiments, and user data through natural language in an AI client.

Description

Amplitude MCP Server is Amplitude's official Model Context Protocol integration. It makes Amplitude analytics, taxonomy, dashboards, experiments, feature flags, cohorts, and session replays accessible from MCP-capable AI clients. According to the official Amplitude documentation, users can query charts, perform ad-hoc segmentation, run funnel and retention analyses, evaluate experiment results, and manage tracking-plan definitions — all without opening the Amplitude UI or writing SQL.

The server runs as a hosted remote service. The official documentation names two regional endpoints: https://mcp.amplitude.com/mcp for the US default and https://mcp.eu.amplitude.com/mcp for EU data residency. Authentication uses OAuth 2.0. The server respects the logged-in user's existing Amplitude permissions: if a user does not have access to a project in Amplitude itself, they cannot retrieve it through an MCP client either.

Analytics and data queries

According to Amplitude's documentation, the MCP Server provides extensive analytics tools. query_chart retrieves data from existing charts, query_experiment analyzes A/B test results including statistical significance, query_dataset executes custom analytics queries, and query_metric returns data for specific metrics. The search tool finds dashboards, charts, notebooks, and experiments through free-text search.

For content access, the source documents get_charts to retrieve full chart definitions, get_dashboard for dashboard content and layout, and get_notebook for notebooks with embedded charts and analyses. get_experiments returns detailed experiment information, and get_flags provides feature-flag configurations.

Taxonomy and tracking-plan management

A key area of the server is tracking-plan management. Amplitude's documentation describes write tools for taxonomy operations: manage_amp_events creates, updates, deletes, and restores events, custom events, and labeled events, and can add newly discovered unexpected events to the plan. create_properties creates event, user, or group properties, and update_properties updates their metadata. manage_amp_data_taxonomy drives the branch workflow for tracking-plan branches: creating, updating, merging, deleting, reviewing, and refreshing branches.

According to the documentation, write operations require the RBAC action "Use MCP (write)". This permission level is controlled by organization administrators, enabling granular access control within MCP access.

Session replay and user insights

get_session_replays enables searching and filtering session recordings based on criteria. This allows finding user sessions that match specific events, errors, or behavioral patterns without opening the Amplitude interface. get_event_properties explores event-property definitions, and get_context returns user and organization information.

The official documentation also describes Progressive Tool Discovery: by appending ?discovery=progressive to the server URL, the client starts with a minimal tool set and discovers tools on demand. This reduces token usage in long-running agent sessions.

Security, data access, and boundaries

All requests through the Amplitude MCP Server leave the local machine and are processed via Amplitude's cloud. The documentation states: "Third-party AI models process your data during analysis." This means the prompt, tool call, result, and database queries can be passed to the selected AI model provider (such as Anthropic, OpenAI, or Google) and to the Amplitude API. Review the privacy and retention policies of both providers before sensitive customer data enters queries.

According to the documentation, the Amplitude MCP Server is exclusively a read and write tool for analytics content — not an event ingestion endpoint. Production events must still be sent through an Amplitude SDK or the HTTP V2 API. The manage_amp_events tool registers event definitions in the tracking plan but does not send event data.

Organization administrators can restrict MCP access through RBAC roles. The documentation describes two separate roles: one for read access and one for write operations. This allows granting specific user groups read-only or full MCP access. The documentation recommends assigning roles deliberately rather than enabling organization-wide access.

The server is in beta according to the README and documentation. Amplitude notes that not all features may be available and that rate limits apply. Plan for appropriate error handling in production-critical analytics workflows.

AI Context and reference files

Amplitude's documentation describes the get_amplitude_context tool: when called with a project ID, it returns the organization-specific AI Context configuration — metric definitions, terminology, and business rules that the user has configured in Amplitude AI Controls. The documentation recommends calling this tool at the start of a session so the AI client works with the correct metrics and terminology.

For repo-side instrumentation conventions, Amplitude documents an optional .amplitude/instrumentation-agent-context.md file at the repository root. Marketplace skills such as instrument-events read this file automatically.

The public GitHub repository amplitude/mcp-server-guide contains only a guide and README; there is no public open-source repository of the actual MCP server. On 2026-09-08, the GitHub API reported exactly 45 stars. This number is a popularity signal for the guide repository, not a quality indicator for the hosted service.

FAQ

Can the Amplitude MCP Server ingest events? No. According to the official documentation, the MCP Server is not an ingestion endpoint. Use an Amplitude SDK or the HTTP V2 API for event tracking.

Is a new Amplitude license required? According to the documentation, any existing Amplitude customer can use the MCP Server from a compatible client. Review current account terms directly with Amplitude.

Can other users access my data? No. The server respects existing Amplitude permissions. Organization administrators additionally control MCP access through RBAC.

Which endpoint is correct for EU data residency? The documentation names https://mcp.eu.amplitude.com/mcp for EU residency.

Requirements

An existing Amplitude account, an MCP-capable AI client with OAuth support, and appropriate project permissions in Amplitude.

Installation instructions

Amplitude MCP Server brings Amplitude analytics, dashboards, experiments, feature flags, and tracking-plan tools into an MCP-capable AI client. The official documentation names two regional remote endpoints: https://mcp.amplitude.com/mcp for the US default and https://mcp.eu.amplitude.com/mcp for EU data residency. No local installation is required — the connection is made through OAuth 2.0 from within the client.

Step 1: Before connecting: clarify the task and permissions

Before setup, make sure you have access to the relevant projects in Amplitude. The MCP Server inherits exclusively the permissions of the authenticated Amplitude user; there is no separate service account. If you want to use write operations — creating or modifying events, properties, branches, or dashboards — an organization administrator must assign you the RBAC action "Use MCP (write)". Read-only access is sufficient for purely analytical tasks.

Think ahead about which data should enter the AI client. According to Amplitude's documentation, third-party AI models process data during analysis. Queries, results, and chart data can therefore be passed to the model provider used by the connected client. For sensitive customer data or regulated information, review the privacy policies of both Amplitude and the selected model provider before including them in queries.

Step 2: Connect the Remote MCP via OAuth

Open the MCP configuration of the chosen client and add the regional server URL. For the US default: https://mcp.amplitude.com/mcp. For EU data residency: https://mcp.eu.amplitude.com/mcp. The client then opens an OAuth 2.0 authorization dialog. Sign in with the Amplitude account that has access to the relevant projects and grant the requested permissions. After successful completion, the MCP Server is active.

For Claude Code, the official documentation describes HTTP transport and a browser-based OAuth flow as an example. Cursor, ChatGPT, and other compatible clients follow their respective MCP configuration steps. Consult the documentation of your specific client for the exact configuration steps.

To reduce token costs, Amplitude documents a Progressive Discovery mode: append ?discovery=progressive to the URL — for example https://mcp.amplitude.com/mcp?discovery=progressive. In this mode, the client starts with a minimal tool set and discovers additional tools on demand via list_tool_categories and get_category_tools.

Use AI Context for better results

The official documentation recommends calling get_amplitude_context with a project ID at the start of each session. This tool returns the configurations stored in Amplitude AI Controls: metric definitions, terminology, and business rules for your organization. When the AI client has this information, it works with the correct terms and metrics in follow-up queries, which reduces misunderstandings about definitions.

Organization administrators can configure AI Context in the Amplitude dashboard under AI Controls: Markdown text instructions and uploaded reference documents. The text instructions are returned by get_amplitude_context; uploaded files are available to in-product Agents but not directly retrievable through MCP tools. Therefore, put rules and definitions that the MCP client should know in text instructions.

Step 3: First queries and typical workflows

Start with a simple search query: "Find all dashboards about onboarding." This tests the connection and shows which projects the client can access. Then try query_chart for a known chart or get_dashboard for an existing dashboard.

A typical analysis workflow according to Amplitude's documentation: search for relevant content with search, retrieve chart definitions with get_charts, run custom queries with query_dataset, and ask follow-up questions based on results. For experiments: get_experiments returns metadata, query_experiment the statistical results.

For Session Replay: get_session_replays with filter criteria such as event name, user attribute, or time range. The result contains matching session IDs and metadata that you can open directly in the Amplitude interface.

Write operations: managing the tracking plan

If you have the RBAC action "Use MCP (write)", you can make tracking-plan changes. Use manage_amp_data_taxonomy to create a branch before changing event definitions — this keeps changes separable and reviewable. manage_amp_events creates new events or updates existing ones. create_properties and update_properties manage properties.

Important: these tools modify tracking-plan definitions in Amplitude Data. They do not send production events. For sending events, use an Amplitude SDK or the HTTP V2 API.

Security, rate limits, and beta status

The Amplitude MCP Server is in beta according to the README and documentation. Rate limits are documented and may change. Plan for appropriate error handling in production workflows and test new features in a test environment first.

Organization administrators can restrict MCP access through AI Controls in the Amplitude dashboard. The documentation describes a separate RBAC role for write operations. Enable write access only for users who genuinely need it. For teams, it is advisable to initially enable only read access and to release write operations deliberately.

Amplitude analytics data — including user segments, event volumes, and experiment results — is sensitive product data. It should not end up in public chats, screenshots, or unauthorized systems. After each setup, check which data the client can actually retrieve and limit queries to what is necessary.

FAQ

Which endpoint is correct for EU data residency? The documentation names https://mcp.eu.amplitude.com/mcp.

Can I use the MCP Server for event tracking? No. According to the official documentation, the MCP Server is not an ingestion endpoint. Use an Amplitude SDK or the HTTP V2 API for production events.

What is Progressive Discovery mode? Appending ?discovery=progressive to the server URL reduces the initially transmitted tool set. The client discovers tools on demand, which reduces token costs.

Do I need a special Amplitude license? According to the documentation, all existing Amplitude customers can use the MCP Server. Review account limits and terms directly with Amplitude.

Authentication

OAuth 2.0. The server inherits the authenticated user's existing Amplitude permissions. Organization administrators control MCP access through RBAC roles.

Required access permissions

Access is limited to the Amplitude projects and data the authenticated user is permitted to see in Amplitude itself. Write access requires the RBAC action "Use MCP (write)".

Transmitted or stored data

Requests and responses are processed through Amplitude's cloud. According to the provider, third-party AI models process data during analysis. Review the privacy and retention policies of both Amplitude and the selected model provider.

Security risks

Amplitude analytics data and user data enter the context of the connected AI client and model. MCP access is controllable through RBAC. Beta status: rate limits and missing features are possible.

License and costs

License
Not recorded yet.
Cost
paid

According to the documentation, the MCP Server is available to all existing Amplitude customers. Costs and account limits depend on the existing Amplitude plan. This entry lists no specific prices; review them directly with Amplitude.

Alternatives

Not recorded yet.

At a glance

Provider
Amplitude
Status
Official server
Deployment
Remote
Current version
Not recorded yet.
GitHub stars
48
Last reviewed
08.09.2026

Repository and documentation

Categories

Supported clients

Related guides

Guides and background related to this entry.

No related guides yet.