Title: Amplitude MCP Server
URL Source: https://amplitude.com/docs/amplitude-ai/amplitude-mcp
Markdown Content:
Use the Amplitude MCP for these jobs:
- Search and read Amplitude content: charts, dashboards, notebooks, experiments, cohorts, events, and properties.
- Query analytics, run ad-hoc segmentation, funnels, and retention, and analyze experiment results.
- Manage taxonomy: create, update, delete, and restore events and properties, manage tracking-plan branches, and read transformations and group types.
- Create and edit objects: charts, dashboards, notebooks, experiments, cohorts, feature flags, and metrics.
- Check data warehouse and cloud storage connections: list import sources and export destinations, and review import and export job history.
Don't use the Amplitude MCP for these jobs:
- Sending production or high-volume event data. The MCP isn't an ingestion endpoint. Emit events with an Amplitude SDK or the HTTP V2 API. The
manage_amp_events tool registers event definitions in the tracking plan; it doesn't stream event data.
- Reading public documentation in an AI client. Use the read-only Docs MCP server instead.
Choose an interface for AI agents
When you build an AI agent that integrates with Amplitude, match each job to the right interface:
| Job | Use | Why |
|---|
| Emit event data (production or high volume) | SDK or HTTP V2 API | The ingestion path. The MCP doesn't ingest events. |
| Validate taxonomy, projects, and queries | Amplitude MCP | Read, search, query, and tracking-plan tools. |
| Create or edit objects (dashboards, experiments, cohorts, charts) | Amplitude MCP tools, or REST APIs in automation | MCP for conversational workflows, REST APIs for scripted pipelines. |
| Read public docs in an AI client | Docs MCP server | Read-only and unauthenticated. |
| Anything the interfaces above don't cover | Browser UI | Use the Amplitude UI for the remaining gaps. |
AI Context and reference files
When an MCP client calls get_amplitude_context with a project ID, the response includes your organization and project markdown AI Context text from AI Controls. Call this tool at the start of a session so the client applies your metric definitions, terminology, and business rules to later queries.
AI Context has two configuration surfaces:
- Text instructions: Markdown you enter in AI Controls. Amplitude MCP returns org and project text when you call
get_amplitude_context with a project ID.
- Uploaded reference files: Documents and spreadsheets you upload in AI Controls. These supplement in-product Agents with detailed reference material. Amplitude MCP tools don't return uploaded file contents. Put rules and definitions the client needs in text instructions instead.
For repo-local instrumentation conventions, add an optional .amplitude/instrumentation-agent-context.md file at your repository root. Amplitude MCP marketplace skills such as instrument-events read this file during instrumentation workflows. This file is separate from AI Controls context.
Configure AI Context in AI Context and AI Controls.
Tracking standards
Amplitude MCP clients read your tracking standard from two places:
- Naming conventions in Data > Settings set event and property casing and, for events, word order. Clients read them with
get_amp_taxonomy (action: workspace). When AI Context text describes a different naming rule, clients follow the settings and tell you about the conflict.
- AI Context text holds the rules the naming settings can't express, such as description wording, categories, required properties, allowed values, and prefixes.
Set casing in Data > Settings, and keep the remaining rules in AI Context text instructions rather than an uploaded file.
Amplitude MCP exposes tools to read and write your project's tracking plan taxonomy. Create, update, delete, and restore calls require the Use MCP (write) RBAC action. Changes to events and properties also require the Update Tracking Plan permission in Amplitude Data. Changes to custom events require the Manage Data Enrichments permission instead.
Read tools
get_amp_taxonomy reads events, properties, the changes on a branch, the activity history of who changed what, and workspace context for a Data project or portfolio: governance settings, environments, and tracking-plan branches. get_transformations and get_group_types cover transformations and group types.
Write tools
manage_amp_events: Create, update, delete, and restore events, custom events, and labeled events, and add ingested unexpected events to the plan.manage_amp_properties: Create event, user, and group properties, and update, delete, and restore event and user properties.manage_amp_branches: Run the branch workflow to create, update, merge, delete, review, and refresh tracking-plan branches.
These tools modify tracking plan definitions in Amplitude Data. They don't send production event data. For the UI-based workflow, refer to Create a tracking plan.
Delete and restore events and properties
Delete and restore tools let an AI client finish a taxonomy cleanup without leaving the conversation. Ask your client to delete redundant events, event properties, or user properties, and to restore anything you deleted by mistake. manage_amp_events handles events and manage_amp_properties handles properties.
What happens to an item depends on whether Amplitude has ingested it:
- Ingested events and properties: Amplitude marks them as deleted and stops ingesting them. You can restore them later with the matching restore action.
- Planned events and properties: Amplitude removes items that you added to the tracking plan but never ingested. You can't restore a removed item. Create it again if you need it.
- Custom and labeled events: Amplitude deletes them permanently, the same as in the Amplitude UI. You can't restore them.
Every delete is a two-step action. The first call returns a confirmation that lists the affected items, project, and branch. Your client relays that summary and asks you to confirm before it calls the tool again with confirmed set to true. Deletes on the main branch take effect immediately. Deletes on any other tracking-plan branch don't affect ingestion until you merge the branch into main.
For an event property, you can delete the property from a single event or plan-wide across every event. Tell your client which scope you mean. A plan-wide delete of a planned property removes it entirely, so ask for the single-event scope when you only want to detach it from one event.
Restoring an event or property on main resumes ingestion right away. On any other branch, Amplitude stages the restore until you merge. To find items you can restore, ask your client to list deleted events or properties. get_amp_taxonomy returns them when you set includeDeleted.
You can't delete or restore Amplitude-owned user properties. Deleting an unexpected event or property that Amplitude has already ingested marks it as deleted, so you can restore it.
| Region | MCP Server URL |
|---|
| United States Server (Default) | https://mcp.amplitude.com/mcp |
| EU Residency Server | https://mcp.eu.amplitude.com/mcp |
Use the Standard Server URL unless your Amplitude data resides in the EU region.
The Amplitude MCP provides comprehensive access to your analytics through these tools:
The tools listed here are the full Amplitude MCP server catalog, available to Cursor, Claude, and most other clients. ChatGPT connects through a curated subset of approximately 25 tools due to a token cap on connected tool schemas. The ChatGPT subset doesn't include session replay tools or progressive discovery tools. To use session replay or progressive discovery, connect with Cursor or Claude instead.
If you connect to Amplitude MCP manually and want to reduce token spend, add ?discovery=progressive to the MCP URL to enable progressive tool discovery.
Progressive discovery tools (list_tool_categories, get_category_tools, describe_tool) aren't part of the ChatGPT tool subset. Progressive discovery requires Cursor, Claude, or another client that receives the full tool catalog.
Progressive URLs (copy/paste):
- US:
https://mcp.amplitude.com/mcp?discovery=progressive.
- EU:
https://mcp.eu.amplitude.com/mcp?discovery=progressive.
In this mode, the initial tool list is intentionally small. The server discovers tools on demand, and all supported tools remain callable after discovery.
Use this workflow:
- Call
get_amplitude_context to confirm organization and project access.
- Call
list_tool_categories to see available product surfaces.
- Call
get_category_tools with a category name to discover tools in that surface.
- Call
describe_tool for any tool you plan to call so you have the latest schema and required parameters.
- Call the tool directly once you confirm the schema.
In progressive mode, tools/list returns discovery tools plus get_amplitude_context first. Use category discovery and describe_tool to find schemas before calling specific tools.
Example progressive discovery flow
Use this sequence in your MCP client conversation:
- "Run
get_amplitude_context and pick the right project."
- "Run
list_tool_categories and show which categories are available."
- "For the
analytics category, run get_category_tools."
- "For
query_amplitude_data, run describe_tool and then call it with valid parameters."
Select your client to open its setup guide.
Analytics and charts
What were the daily active users over the last 7 days?
Show signup conversion rates by traffic source this month.
Create a chart showing weekly active users broken down by platform.
Why did this metric spike last week? [paste chart URL]
Dashboards and notebooks
Build an executive dashboard for our product launch.
Summarize this dashboard for my exec meeting. [paste dashboard URL]
Add a retention chart to the KPI dashboard.
Remove the DAU chart from the overview dashboard.
Update the title of the KPI notebook to "Q2 Growth Metrics".
Move the funnel chart to the top of the onboarding dashboard and add a section header above it.
Experiments and feature flags
Create an experiment to test the new checkout flow with a 50/50 split.
Should we ship the new onboarding experiment? Analyze the results.
Check on all running experiments and flag anything stale or missing metrics.
Create a feature flag for the redesigned settings page.
Session Replay and debugging
Show me session replays where users rage-clicked on the checkout page.
A user reported a bug on the billing page. Find their session and walk me through what happened.
Audit the onboarding flow for UX friction using session replays.
Feedback and opportunities
What are the top customer complaints this month?
How has feedback about the mobile app trended over the last 90 days?
Find me the biggest product opportunities. Where are we losing users?
AI agent analytics
How are our AI agents performing? Show quality, cost, and error rates.
What topics are users asking the AI assistant about?
Investigate the low-quality AI sessions from yesterday. What went wrong?
Build a dataset from last week's failed agent sessions and run the refund-policy evaluator against it.
Taxonomy and tracking plan
What branches exist in my project?
Create a branch called "q3-cleanup" for tracking plan changes.
Create an event called "Checkout Started" on the q3-cleanup branch.
Add unexpected events "app_opened" and "checkout_completed" to the tracking plan.
Create a user property called "plan_type" for my project.
Delete the events "old_checkout" and "legacy_signup" from the tracking plan.
Remove the "page_url" property from the "Page Viewed" event only.
Restore the deleted user property "plan_type".
Merge the q3-cleanup branch into main.
Guides and Surveys
What are the published guides with the highest priority in my project?
Which guides are live on the pricing page right now?
Show me the full targeting and trigger configuration for guide 15731.
Users, cohorts, and metrics
Create a cohort of power users who completed 10+ sessions in the last 30 days.
Create a "revenue per user" metric for the checkout experiment.
Compare user behavior between converters and drop-offs in the onboarding flow.
Sync my "High-intent buyers" cohort to Iterable.
Data warehouse
What warehouse import sources are configured for my project?
Show me the recent import jobs for my Snowflake source. Which ones failed or completed with errors?
Where does this project export its data, and when did each destination last succeed?
Show me the manual backfill export jobs for my Snowflake destination.
Analysis workflow
Most analysis questions follow this pattern:
- Search: Find relevant charts, dashboards, or experiments.
- Retrieve: Get full definitions and metadata.
- Query: Execute data queries and return results.
- Analyze: Ask follow-up questions and dive deeper.
Example conversation:
You: Why did weekly active users drop last Tuesday?AI: [Searches for WAU charts, queries the data, identifies the drop]AI: Looks like WAU dropped 12% on Tuesday. Let me check if any experiments or deployments landed that day.AI: [Cross-references deployments and experiment launches]AI: A deployment shipped Monday evening that changed the session timeout. That likely caused the drop.You: Show me session replays from affected users.AI: [Finds replays, extracts interaction timelines, summarizes friction patterns]
Creation and editing workflow
You can create and edit Amplitude content directly through conversation:
- Create: Build new charts, dashboards, notebooks, experiments, cohorts, flags, or metrics.
- Configure: Specify parameters and settings conversationally.
- Refine: Edit and adjust layouts, metrics, or variants. For dashboards and notebooks, use
use_amp_dashboards or use_amp_notebooks to add or remove charts, reorder content, and update metadata.
- Deploy: Save and use the content in Amplitude.
Example conversation:
You: Set up an A/B test for the new pricing page.AI: [Creates a "revenue per user" metric and a "checkout completion" guardrail metric]AI: [Creates the experiment with control and variant, attaches both metrics]You: Add the product team as testers so they can preview it.AI: [Updates the flag with testers for the variant]You: Build a dashboard so we can monitor results.AI: [Creates a dashboard with experiment results, conversion funnel, and revenue charts]
The Amplitude MCP Marketplace is the official open-source plugin repository for Amplitude MCP. It provides a curated set of reusable skills, agents, and commands that extend what you can do with the Amplitude MCP server in Claude Code, Cursor, Claude, and Codex.
Available skills
The skills table uses seven areas: core analytics, product insights, session replay and debugging, AI agent analytics, analytics instrumentation, briefings, and bonus.
Core analytics
| Skill | Description |
|---|
| create-chart | Create Amplitude charts from natural language, with automatic event discovery, filters, groupings, and visualization. |
| create-dashboard | Build dashboards from requirements or goals, organizing charts into logical sections with appropriate layouts. |
| analyze-chart | Analyze a specific chart in depth to explain trends, anomalies, and likely drivers. |
| analyze-dashboard | Synthesize an entire dashboard into talking points, surface concerns, and connect quantitative data to qualitative insights. |
Product insights
| Skill | Description |
|---|
| analyze-experiment | Design A/B tests, monitor running experiments, and interpret results with statistical rigor. |
| monitor-experiments | Triage all active and recently completed experiments by importance. |
| analyze-feedback | Synthesize customer feedback into grouped themes such as feature requests, bugs, pain points, and praise. |
| analyze-account-health | Summarize B2B account health with usage patterns, risk signals, and expansion opportunities. |
| discover-opportunities | Find product opportunities by cross-referencing analytics, experiments, replays, and feedback. |
| compare-user-journeys | Compare two user groups side-by-side to surface behavioral differences. |
Session Replay and debugging
| Skill | Description |
|---|
| debug-replay | Turn bug reports into numbered reproduction steps by extracting the interaction timeline from Session Replay. |
| replay-ux-audit | Watch multiple session replays for a flow and synthesize a ranked friction map. |
| diagnose-errors | Triage product issues across network failures, JavaScript errors, and error clicks. |
| monitor-reliability | Generate a proactive reliability report from auto-captured error data so issues surface before users report them. |
AI agent analytics
| Skill | Description |
|---|
| analyze-ai-topics | Analyze what users ask AI agents about and how well the agent answers each topic. |
| investigate-ai-session | Deep-dive specific AI agent sessions or failure patterns for root-cause analysis. |
| monitor-ai-quality | Deliver a proactive health report on AI agents covering quality, cost, performance, and errors. |
| review-agent-insights | Retrieve, synthesize, and prioritize all recent AI agent results across agent types into a unified narrative. |
Analytics instrumentation
| Skill | Description |
|---|
| diff-intake | Read a PR or branch diff and output a structured change brief for analytics planning. |
| discover-event-surfaces | List candidate analytics events from a change brief for PM prioritization. |
| discover-analytics-patterns | Map how analytics is already implemented in the codebase (SDK calls, naming, imports). |
| instrument-events | Build a concrete instrumentation plan and JSON tracking plan from prioritized event candidates. |
| add-analytics-instrumentation | End-to-end workflow: read code, decide what to track, and produce a full instrumentation plan in one pass. |
A typical flow runs diff-intake → discover-event-surfaces → instrument-events, with discover-analytics-patterns to align new tracking with existing conventions.
Briefings
| Skill | Description |
|---|
| daily-brief | Morning briefing of the most important changes across your Amplitude instance. |
| weekly-brief | Weekly recap of trends, wins, and risks to share with your team or leadership. |
Bonus
| Skill | Description |
|---|
| taxonomy | Generate, validate, audit, and score event tracking plans, naming conventions, and data governance best practices. |
| what-would-lenny-do | Answer product strategy questions by searching Lenny Rachitsky's archive (requires lennysdata MCP server). |
Skills activate automatically based on your request. For example, asking "Why did this metric spike last week?" triggers the analyze-chart skill, and asking "Summarize this dashboard for my exec meeting" triggers analyze-dashboard.
Data access
- The MCP server uses your existing Amplitude user permissions and access controls.
- You can only access Amplitude projects and data that you already have permission to view in your regular Amplitude account.
- You receive no additional data access beyond your current Amplitude account privileges.
- OAuth authentication ensures secure connection between the MCP server and your Amplitude account.
- Enterprise-managed auth connects Claude through your identity provider and skips the consent screen.
Privacy considerations
The AI service you're using (for example, Claude or Gemini) processes your Amplitude data. Review your organization's policies regarding AI-powered data analysis tools and consider compliance requirements (General Data Protection Regulation, California Consumer Privacy Act).
Third parties (for example, Anthropic) develop and maintain the AI models used with this MCP server. Amplitude isn't responsible for model outputs, including hallucinations, inaccuracies, or errors resulting from model behavior, even if such outputs use your Amplitude data.
Admin controls
Amplitude enables MCP server access by default for all users in your organization. Organization administrators can opt out or restrict access if needed:
- Navigate to Settings > Content Access > MCP in your Amplitude organization settings.
- Use the content access controls to allow or block the Amplitude MCP server.
- Access restrictions apply to all MCP clients across all users in your organization.
MCP access controls are an admin-only setting. Individual users can't override organization-level MCP restrictions.
Scope MCP access with roles
For finer-grained control than the org-level toggle, two RBAC actions govern what the MCP server can do on a per-project basis: Use MCP (read) (USE_MCP_READ) and Use MCP (write) (USE_MCP_WRITE). Assign these actions to a role, then attach the role to users, groups, or service accounts to scope exactly what an agent can read from or write to.
- Project-level control: restrict a role to read-only or no access on specific projects. An agent running as that user can't write to, or even read from, projects where the role doesn't grant the action.
- Groups and service accounts: the same actions apply to permission groups and to Amplitude MCP service accounts, so team-wide policies and automated agent workflows get the same controls as individual users.
- Mixed access: within a single session, a tool call such as
manage_amp_events succeeds for a project where your role grants read access and fails for a project where it doesn't.
- Applies to everyone: these actions override admin status. If a role doesn't grant the action, MCP calls fail even for an org administrator.
- Enabled by default: existing roles already include these actions. Every role grants read access, and the Member, Manager, and Admin roles also grant write access, so nothing changes until an admin adjusts a role.
To scope MCP access with a role:
- In Org Settings > Role Management, create or edit a role.
- In the AI Features category, enable Use MCP (read) and Use MCP (write).
- Assign the role to a user, group, or service account, scoped to the relevant projects.
Amplitude enforces these actions on each MCP tool call, project by project. The tool list stays visible to all clients, so a call to a tool the role doesn't permit fails at call time rather than disappearing from the list.
Common issues
Authentication and OAuth issues
- Ensure your Amplitude account has proper project access.
- Check that you have logged into the correct Amplitude account.
- Make sure you're only logged into one Amplitude organization during the OAuth flow. Logging into multiple organizations can cause authentication issues.
- Try disconnecting the MCP connection and re-authenticating through the OAuth flow.
- Try logging out of Amplitude, then reconnecting.
- Desktop apps may require restart after configuration changes.
- Authorization page may appear to spin indefinitely (close tab after authentication).
Missing data
- Verify you have access to the specific Amplitude project.
- Check if the data exists in your Amplitude interface.
- Ensure proper permissions for the requested data.
Chart query issues
- AI platforms may truncate some large charts.
- Querying charts from dashboards may use default chart settings instead of saved dashboard filters.
MCP client issues
- Cursor tool call failures: If MCP tool calls fail in Cursor, this is often due to expired or corrupted authentication tokens. Open the Command Palette (
Cmd+Shift+P on Mac, Ctrl+Shift+P on Windows/Linux), type and select "Clear All MCP Tokens," then re-authenticate with your accounts.
- Token limit errors: If you receive token limit exceeded errors, try starting a new conversation thread or increase the maximum token limit in your MCP client settings.
- Connection timeouts: If queries are timing out, try breaking down complex requests into smaller, more focused questions.
- Tool loading failures: If tools aren't loading, restart your MCP client application and re-authenticate.
Getting help
If you encounter issues not covered here:
- Verify your setup matches the configuration examples.
- Test with a simple query like "What Amplitude projects are accessible?"
- Check that your Amplitude account has the necessary permissions.
- Contact your Amplitude administrator for organization-specific setup help.
Amplitude is constantly improving the Amplitude MCP server and would appreciate hearing from you. Share your feedback, suggestions, or report issues using this feedback form.
Transport Type: Streaming HTTP (Remote).
Authentication: OAuth 2.0 with Amplitude, or enterprise-managed auth for Claude through your identity provider.