product-analytics-mcp

Amplitude Product Analytics MCP

Purpose

Connect an AI assistant or coding agent to Amplitude's hosted MCP server so it can analyze product data, retrieve saved content, query charts, inspect experiments, create Amplitude objects, and debug user behavior from the agent's normal workspace.

Default hosted server:

https://mcp.amplitude.com/mcp

EU residency server:

https://mcp.eu.amplitude.com/mcp

Use the US server unless the user's Amplitude data resides in the EU region. The server uses OAuth and runs as the authenticated Amplitude user, so available organizations, projects, dashboards, experiments, replays, and write actions are limited by that user's Amplitude permissions.

When to Use

• Answer questions about Amplitude charts, dashboards, notebooks, cohorts, metrics, feature flags, experiments, events, properties, or product usage.
• Investigate trends, spikes, drops, funnel conversion, retention, segmentation, or experiment results from Amplitude data.
• Create or edit Amplitude charts, dashboards, notebooks, cohorts, metrics, experiments, or feature flags from a natural-language brief.
• Pull Session Replay timelines to debug a user report, rage clicks, errors, or onboarding friction.
• Analyze customer feedback, product opportunities, or AI agent analytics captured in Amplitude.
• Configure Amplitude MCP in Codex CLI, Claude Code, Cursor, ChatGPT, Claude, Gemini CLI, Lovable, Kiro, Replit, Figma Make, or another MCP-capable client.

Do not use this skill for general web analytics if the user has not asked to use Amplitude or if no Amplitude MCP connection/account is available. In that case, ask for the right analytics source or provide setup instructions.

Setup

1. Pick the region

If the user is unsure, start with the default US URL. If OAuth succeeds but expected projects or data are missing, ask whether their Amplitude organization uses EU residency.

2. Configure the MCP client

Use the client's native MCP setup path when available:

For EU residency, replace the URL with https://mcp.eu.amplitude.com/mcp.

3. Authenticate and verify access

After adding the server:

1. Complete the Amplitude OAuth flow.
2. Ensure the MCP tools appear in the client.
3. Call get_context to confirm the authenticated user, organization, and accessible projects.
4. If the client supports project selection, pick the project that matches the user's request.
5. Call get_project_context before project-specific work so time zone, currency, session definitions, and AI context are available.

If no tools appear after OAuth, restart or reload the MCP client. Many clients only load MCP servers at startup.

Progressive Tool Discovery

For clients with tight context budgets, use progressive discovery:

https://mcp.amplitude.com/mcp?discovery=progressive

EU:

https://mcp.eu.amplitude.com/mcp?discovery=progressive

Progressive mode starts with a small tool list, then discovers schemas on demand. Use this sequence before calling a specialized tool:

1. Call get_context to confirm organization and project access.
2. Call list_tool_categories to see available product surfaces.
3. Call get_category_tools for the relevant category, such as analytics, experiments, dashboards, session_replay, feedback, or agent_analytics.
4. Call describe_tool for the exact tool you plan to use.
5. Call the tool only after confirming required parameters and schema.

In progressive mode, do not assume hidden tools are unavailable just because tools/list is short. Discover the relevant category first.

Tool Selection

Common Amplitude MCP tools include:

Tool schemas may change. In progressive mode, always call describe_tool before a first call. Outside progressive mode, inspect the client-visible tool schema before unfamiliar write operations.

Workflows

Analyze a trend or anomaly

1. Search for relevant saved charts, dashboards, metrics, or events.
2. Retrieve full definitions with get_charts, get_dashboard, get_from_url, or get_project_context.
3. Query the data with query_chart, query_charts, or query_amplitude_data.
4. Segment by likely drivers such as platform, traffic source, plan, country, experiment exposure, or feature flag.
5. Cross-check deployments, experiments, session replays, or feedback when the metric movement needs explanation.
6. Return the relevant Amplitude links, date range, project, filters, and concrete data points used.

Example prompts:

What were daily active users over the last 7 days?
Why did this funnel conversion drop last Tuesday? [paste chart URL]
Compare onboarding completion between converters and drop-offs this month.

Create or edit Amplitude content

Write workflows can create or modify real Amplitude objects. Before writes, confirm the target project and summarize the intended change.

1. Search for existing content first so you do not duplicate a dashboard, metric, cohort, or flag.
2. Retrieve current definitions when editing existing objects.
3. Build the smallest valid payload from verified requirements.
4. Prefer temporary chart renders or draft-like intermediate states when available.
5. Before final save, show the title, project, date range, events, filters, chart type, destination dashboard/notebook, and any write side effects.
6. After saving, return the resulting Amplitude URL or object ID.

Safe examples:

Create a chart showing weekly active users broken down by platform.
Build an executive dashboard for launch health using activation, retention, and error signals.
Create a cohort of power users who completed 10 or more sessions in the last 30 days.

For experiments and feature flags, be stricter: confirm rollout, variants, traffic allocation, metrics, and ownership before creating or updating anything.

Debug with Session Replay

1. Identify the user, account, event, session, time range, or replay URL.
2. Use get_session_replays or list_session_replays to find candidate recordings.
3. Use get_session_replay_events to extract the processed interaction timeline.
4. Summarize steps, friction, errors, rage clicks, and likely reproduction paths.
5. Link the replay and note any limits or missing context.

Do not expose private user data unnecessarily. Return only the details needed for debugging.

Analyze feedback or opportunities

Use feedback tools when the user asks about complaints, feature requests, pain points, praise, opportunities, account health, or customer sentiment:

1. Use get_feedback_sources if source filtering matters.
2. Use get_feedback_insights, get_feedback_comments, or get_feedback_trends to gather evidence.
3. Group themes by user impact, frequency, revenue/account relevance, and recency.
4. Cross-reference charts, cohorts, experiments, or replays when the recommendation depends on behavior.

Analyze AI agent quality

Use agent analytics tools when the user's product captures AI agent sessions in Amplitude:

1. Call get_agent_analytics_schema to learn available agents, topics, tools, quality rubrics, and filter fields.
2. Query metrics or sessions by quality, cost, latency, error rate, user sentiment, topic, or agent.
3. Inspect conversations or spans only when needed for root cause.
4. Return aggregate patterns first, then representative examples.

Amplitude MCP Marketplace Plugin

Amplitude also maintains an MCP Marketplace plugin with reusable skills, agents, and commands for Claude Code, Cursor, Claude, and Codex.

Install in clients that support plugin marketplaces:

/plugin marketplace add amplitude/mcp-marketplace
/plugin install amplitude@amplitude

Use the plugin when the user wants higher-level Amplitude workflows such as creating charts, analyzing dashboards, debugging replays, monitoring experiments, planning instrumentation, producing daily or weekly briefs, or reviewing AI agent insights. The plugin complements the MCP server; it does not replace OAuth or the hosted MCP connection.

Security and Compliance

• The MCP server authenticates with OAuth 2.0 and runs as the authenticated Amplitude user.
• Respect organization and project permissions. If a tool cannot see content, do not work around it with unrelated accounts.
• Amplitude admins can enable, block, or restrict MCP access from Settings > Content Access > MCP.
• Individual users cannot override organization-level MCP restrictions.
• The AI client processes Amplitude data after tool calls. Follow the user's organization policy for AI-powered data analysis, GDPR, CCPA, and other compliance requirements.
• Do not paste Amplitude OAuth tokens, cookies, API keys, raw user profiles, unnecessary PII, or private dashboard contents into public logs, PRs, screenshots, or issues.
• Treat third-party feedback and session replay content as sensitive user data.

Troubleshooting

Expected Output

For analysis:

{
  "success": true,
  "project": "Production",
  "date_range": "last_7_days",
  "tools_used": ["search", "query_chart"],
  "summary": "DAU increased 8.4% week over week, driven mostly by mobile users.",
  "evidence": [
    {
      "label": "DAU chart",
      "url": "https://app.amplitude.com/analytics/example/chart/abc123",
      "metric": "daily_active_users",
      "value": 12345
    }
  ],
  "next_steps": ["Segment by acquisition source", "Review onboarding replays for new mobile users"]
}

For setup:

{
  "success": true,
  "server_url": "https://mcp.amplitude.com/mcp",
  "client": "codex",
  "verification": "get_context returned the expected organization and project",
  "notes": ["Use the EU URL only for EU-residency organizations"]
}