Amplitude MCP use cases
These example prompts and multi-step workflows cover common tasks available through Amplitude MCP after you connect a client. Start with a question, name your project and date range, and add relevant events, metrics, or content links as you refine the results. Your agent selects the appropriate tools from the tool catalog automatically.
For a task-to-tool lookup that maps each job to the tools the server runs, refer to Common tasks.
Example prompts
Choose a starting point and adapt it to your project. Results depend on your data, product access, and permissions.
Analytics and charts
What were the daily active users over the last 7 days?
Expected result: Daily active user counts for the last seven days, with the activity definition, date range, and project time zone used.
Show signup conversion rates by traffic source this month.
Expected result: Signup conversion rates by traffic source for the current month, including the funnel events, conversion window, and number of users in each segment.
Create a chart showing weekly active users broken down by platform.
Expected result: A weekly active users chart grouped by platform, with its event definition and date range, and a link or preview to review.
Why did this metric spike last week? [paste chart URL]
Expected result: A comparison with the preceding period, segments contributing to the spike, and possible explanations supported by the available evidence.
Explore a metric
In our production project, compare weekly signup conversion for the last four complete weeks. Confirm the signup and completion events first, and show the definition and time zone used.
Expected result: A week-by-week comparison of signup conversion, with the confirmed funnel events, metric definition, date range, and time zone used.
Dashboards and notebooks
Build an executive dashboard for our product launch.
Expected result: A launch dashboard with agreed success metrics and relevant charts, organized for an executive audience, plus its Amplitude link.
Summarize this dashboard for my exec meeting. [paste dashboard URL]
Expected result: An executive summary of the dashboard's trends, notable changes, and open questions, with links to the supporting charts.
Add a retention chart to the KPI dashboard.
Expected result: The selected KPI dashboard updated with a retention chart, with the chart definition and updated dashboard link.
Remove the DAU chart from the overview dashboard.
Expected result: The DAU chart removed from the identified dashboard's layout, with a link to the updated dashboard. The underlying saved chart remains available.
Update the title of the KPI notebook to "Q2 Growth Metrics".
Expected result: The selected notebook renamed to "Q2 Growth Metrics", with its updated title and link.
Move the funnel chart to the top of the onboarding dashboard and add a section header above it.
Expected result: The onboarding dashboard reordered with the funnel chart at the top and a section header above it, plus the updated dashboard link.
Find existing work
Find dashboards about onboarding. Summarize the three most relevant ones and link to the underlying charts.
Expected result: Up to three relevant onboarding dashboards, a short summary of each, and links to the dashboards and underlying charts.
Experiments and feature flags
Create an experiment to test the new checkout flow with a 50/50 split.
Expected result: A feature experiment configured with control and variant at a 50/50 split, with the chosen metrics, targeting, and experiment link for review.
Should we ship the new onboarding experiment? Analyze the results.
Expected result: A recommendation based on variant performance, primary and guardrail metrics, statistical uncertainty, and any limits that affect the decision.
Check on all running experiments and flag anything stale or missing metrics.
Expected result: A review of accessible running experiments, highlighting missing metrics and experiments that need attention, with reasons and links.
Create a feature flag for the redesigned settings page.
Expected result: A feature flag for the redesigned settings page, with its key, variants, targeting settings, and link for review before rollout.
Understand an experiment
Analyze the checkout experiment at this Amplitude URL. Report the variants, primary metric, observed results, and statistical uncertainty. Keep the experiment unchanged.
Expected result: An experiment readout covering the variants, primary metric, observed performance, and statistical uncertainty to support your decision.
Session Replay and debugging
Show me session replays where users rage-clicked on the checkout page.
Expected result: Matching checkout replays using available rage-click or interaction filters, with a summary of the recorded behavior and links to review each recording.
A user reported a bug on the billing page. Find their session and walk me through what happened.
Expected result: A matching session after you provide the user and approximate time, with a chronological summary of billing-page interactions, relevant events, and a replay link.
Audit the onboarding flow for UX friction using session replays.
Expected result: Recurring friction patterns observed in the sampled onboarding replays, with supporting timelines, recording links, and suggested follow-up.
Investigate customer friction
Find recent session replays with repeated checkout attempts. Summarize their interaction events and timelines, and link to recordings I can review.
Expected result: A shortlist of matching replays, a summary of the checkout interactions and timelines, and links to recordings for review.
Feedback and opportunities
What are the top customer complaints this month?
Expected result: Recurring complaint themes from connected feedback sources for the current month, with representative comments and the evidence behind their relative frequency.
How has feedback about the mobile app trended over the last 90 days?
Expected result: Changes in mobile-app feedback themes over the last 90 days, including recurring issues, emerging concerns, and representative supporting comments.
Find me the biggest product opportunities. Where are we losing users?
Expected result: Product opportunities grounded in funnel drop-offs, retention patterns, and available feedback or replays, with the affected segments, supporting evidence, and suggested investigations.
Review customer feedback
Find feedback themes about onboarding. Show representative comments and distinguish frequently mentioned problems from isolated reports.
Expected result: A summary of onboarding feedback themes with representative comments and a clear distinction between recurring issues and isolated reports.
AI agent analytics
How are our AI agents performing? Show quality, cost, and error rates.
Expected result: An agent performance summary using the quality, cost, and error fields available in your schema, with metric definitions, date range, and any data gaps.
What topics are users asking the AI assistant about?
Expected result: Common conversation topics found in accessible AI-assistant sessions, with representative examples and their relative prevalence in the reviewed data.
Investigate the low-quality AI sessions from yesterday. What went wrong?
Expected result: A review of sessions matching the available quality signals, with recurring failures supported by transcripts or operation spans and recommended follow-up.
Build a dataset from last week's failed agent sessions and run the refund-policy evaluator against it.
Expected result: A dataset of matching failed sessions and an evaluation run using an available refund-policy evaluator, with dataset and run references and the results available to your agent.
Dataset and evaluation workflows use run_amp with the agent-analytics commands and require Agent Analytics access and a configured evaluator.
Understand AI-agent behavior
Compare tool reliability and latency across our AI-agent sessions last week. Describe the available schema before choosing the analysis.
Expected result: A comparison of tool reliability and latency across last week's AI-agent sessions, with the schema fields and metric definitions used in the analysis.
Taxonomy and tracking plan
What branches exist in my project?
Expected result: The project's tracking-plan branches, with available names, identifiers, and status information.
Create a branch called "q3-cleanup" for tracking plan changes.
Expected result: A tracking-plan branch named "q3-cleanup", with its identifier and a summary of its starting state.
Create an event called "Checkout Started" on the q3-cleanup branch.
Expected result: A "Checkout Started" event definition on the q3-cleanup branch, with its details and the branch where it was created.
Add unexpected events "app_opened" and "checkout_completed" to the tracking plan.
Expected result: The two ingested events added to the selected tracking-plan branch, with their definitions and resulting plan status.
Create a user property called "plan_type" for my project.
Expected result: A user-property definition named "plan_type" in the project's tracking plan, with the chosen type and description.
Delete the events "old_checkout" and "legacy_signup" from the tracking plan.
Expected result: A preview of the affected event definitions and required confirmation, followed by the deletion status for each event after approval.
Remove the "page_url" property from the "Page Viewed" event only.
Expected result: The property removed at the "Page Viewed" event scope after any required confirmation, with the affected scope explicitly reported.
Restore the deleted user property "plan_type".
Expected result: The eligible deleted user property restored, with its resulting tracking-plan status or an explanation if it can't be restored.
Merge the q3-cleanup branch into main.
Expected result: A branch-difference review and any conflicts or required confirmation, followed by a merge summary after approval.
Tracking-plan changes update definitions. Review the taxonomy and delete/restore notes before deleting or restoring items.
Guides and Surveys
Use run_amp to inspect guides and surveys and work with unpublished drafts. Publishing, audience configuration, scheduling, and testing remain in Amplitude. Guides and Surveys reads are unavailable through enterprise-managed auth.
What are the published guides with the highest priority in my project?
Expected result: Published guides ordered by their configured priority, with names, identifiers, and available targeting details.
Which guides are live on the pricing page right now?
Expected result: Published guides whose page-targeting rules match the pricing page, with their trigger and audience conditions. Individual users must also meet those conditions to see a guide.
Show me the full targeting and trigger configuration for guide 15731.
Expected result: The guide's available page, audience, and trigger configuration, with its publication status and identifier.
Show me all guides created by alice@example.com in Q1 2025 in project 12345.
Expected result: Matching guides using the supported creator and creation-date filters, with identifiers, available metadata, and pagination when the results span more than one page.
List all non-archived guides and surveys in project 12345. Group them by rollout state so I can see what is published, testing, scheduled, or in an experiment.
Expected result: A paginated inventory of non-archived items grouped by available rollout status, including experiment-running items when present and available schedule details. Published status alone does not establish that every user is eligible to see an item.
Which iOS surveys are currently published in project 12345?
Expected result: Published iOS surveys using the platform filter and available rollout-state information, with identifiers and admin links. Delivery still depends on the individual user's targeting and trigger conditions.
Which web guides were created in the last 30 days in project 12345?
Expected result: Web guides matching the supported platform and creation-date filters, with titles, identifiers, creation metadata where available, and pagination as needed.
List all archived surveys in project 12345.
Expected result: Archived surveys identified from results that include archived items, with their identifiers and available metadata. The agent follows pagination rather than treating the first page as the full inventory.
Find guide 15731 in project 12345 and show its full published configuration: page targeting, triggers, copy in each step, and snooze or dismiss settings.
Expected result: The published configuration for the selected guide, including available targeting and lifecycle settings, step content, variants, identifier, and admin link. The agent uses a listing first if it needs to resolve the item; reading the unpublished draft requires the separate draft option.
Read the latest unpublished draft for guide 15731 in project 12345. Show its steps, content variants, page targeting, snooze and dismiss settings, and the version needed for an update.
Expected result: The latest draft configuration retrieved with include-unpublished-changes, including available content and lifecycle settings and latest_version. The published version remains unchanged.
Rename survey 678 in project 12345 to "NPS Q4" and raise its priority to urgent. Read the latest draft version first and leave the changes unpublished.
Expected result: A draft metadata update using the version read from the latest draft. If the version is stale, the agent reports the conflict and reads the current version before proposing a retry. Live users continue to see the published version.
Check whether guide 15731 in project 12345 can be archived. Use a dry run and do not archive it yet.
Expected result: Validation without an archive mutation, or an explanation that the item is still published and must be unpublished in Amplitude first. A later archive requires destructive-operation opt-in.
Create an unpublished guide called "Billing refresh announcement" in project 12345 with a single modal step. Return its admin URL so I can set the audience, preview it, and publish in Amplitude.
Expected result: An unpublished guide on a disabled draft flag and its admin URL. Audience selection, preview, publishing, scheduling, and testing remain in Amplitude. If the write result is ambiguous, the agent checks the existing items before retrying.
Users, cohorts, and metrics
Create a cohort of power users who completed 10+ sessions in the last 30 days.
Expected result: A dynamic cohort defined by at least 10 sessions in the last 30 days, with the session definition, membership criteria, and cohort link.
Create a "revenue per user" metric for the checkout experiment.
Expected result: A reusable revenue-per-user metric using the agreed revenue event or property and user denominator, with its definition and reference for the experiment.
Compare user behavior between converters and drop-offs in the onboarding flow.
Expected result: A comparison of the two populations' event patterns and relevant properties, with the funnel definition, sample sizes, and notable differences.
Sync my "High-intent buyers" cohort to Iterable.
Expected result: A preview of the cohort sync to a configured Iterable destination, followed by the sync job status after you confirm it.
Data warehouse
Data warehouse workflows depend on your region, organization access, and connected warehouse. Availability isn't yet uniform across US and EU.
What warehouse import sources are configured for my project?
Expected result: Where available, the project's configured warehouse import sources, with their identifiers, source types, and available status details.
Show me the recent import jobs for my Snowflake source. Which ones failed or completed with errors?
Expected result: Where available, recent jobs for the selected Snowflake source, highlighting failed jobs and jobs with errors, with timestamps and available error details.
Where does this project export its data, and when did each destination last succeed?
Expected result: Where available, configured export destinations and their most recent successful exports, including any destinations without a recorded success.
Show me the manual backfill export jobs for my Snowflake destination.
Expected result: Where available, manual backfill export jobs for the selected destination, with their date ranges, execution status, and available error details.
What data sources are configured for my project? Include SDKs and integrations as well as warehouse imports, and identify any Snowflake or BigQuery sources.
Expected result: Where available, the project's ingestion sources, including SDKs, integrations, and warehouse imports, with identifiers, source types, status, and recent volume when reported. Sources can then be selected by name or type for follow-up.
Show the configuration for my Snowflake ingestion source, including its schedule and any available schema details.
Expected result: The selected ingestion source's configuration and available metadata, resolved from the source listing. Export-destination configuration is outside the source-detail tool's scope.
How many records did the last BigQuery import bring in? Tell me whether Amplitude reported a count or the count is unavailable.
Expected result: Where available, the latest job's recordCount and recordCountReported. A reported zero is distinguished from an unavailable count, with available row-error information.
Check whether my Snowflake export destination delivered data recently. Compare its automatic export jobs with its manual or backfill jobs, and explain any difference between the latest successful job time and the latest export completion time.
Expected result: Where available, destination timestamps and separate automatic/manual job histories. The explanation distinguishes a successful zero-row automatic job from a run that exported data and notes that successful zero-row automatic exports are omitted from job history.
Workflows
Use a sequence of prompts when a task needs exploration, review, and a saved result.
Analyze product performance
Provide: your project, business question, date range, and any relevant chart, dashboard, or experiment links.
Find charts, dashboards, and experiments relevant to weekly active users. Read their definitions and metadata, then identify which content best answers our question and explain the date range, filters, and activity definition.
Expected result: Relevant Amplitude content and links, with the definitions and context needed to choose the right analysis.
Query the relevant charts to compare weekly active users across the selected periods. Identify the largest changes and contributing segments. Use available experiment launches and the deployment context I provide to suggest explanations, and distinguish observations from hypotheses.
Expected result: A sourced comparison of weekly active users, contributing segments, and possible explanations with their evidence limits.
Find session replays from the affected users and inspect their recorded interactions and event timelines. Summarize recurring friction and explain whether the replay evidence supports the leading hypotheses.
Expected result: Supporting replay links and interaction summaries, with a clear account of what the evidence supports and what still needs verification.
Create and edit content
Provide: your project, the change you want to test, metric definitions, targeting requirements, and the teammates who need preview access.
Propose a feature experiment for the new pricing page with control and variant groups, a revenue per user primary metric, and a checkout completion guardrail metric. Find existing metrics first and show me the definitions and configuration before creating anything.
Expected result: A proposed feature experiment and primary and guardrail metric definitions, with the targeting and allocation ready for review.
Create the feature experiment using the approved configuration and metrics. Add the selected product-team members as testers for the variant so they can preview it. Return the experiment and flag links for review before starting the experiment.
Expected result: A saved feature experiment with control and variant groups, attached metrics, and the selected testers, plus links to review its configuration.
Build a dashboard to monitor this experiment with experiment results, a conversion funnel, and revenue charts. Add a description explaining the metric definitions and date ranges. Return the dashboard link so we can review and refine its layout.
Expected result: A saved monitoring dashboard with the experiment's results, funnel, and revenue charts, with clear definitions and an Amplitude link for review.
Investigate a metric change
Provide: your project, a metric definition or chart URL, comparison periods, and relevant release or campaign dates.
Read this chart and compare the last complete week with the prior week. Explain the metric definition, filters, sample sizes, and size of the change.
Expected result: A comparison of the two complete weeks, with the metric definition, filters, sample sizes, and absolute and relative changes.
Break the change down by device, acquisition channel, and customer segment where the schema supports them. Separate observed contributions from hypotheses.
Expected result: A breakdown of the change by supported segments, distinguishing measured contributions from possible explanations.
Inspect relevant user activity, replay timelines, or feedback to test the strongest hypotheses. Summarize what the data supports and what remains uncertain.
Expected result: Supporting user activity, replay timelines, or feedback for the strongest hypotheses, with remaining evidence gaps.
Build and refine a dashboard
Provide: your project, audience, business question, and the definitions your team uses.
Propose an onboarding dashboard with activation, time to value, and retention. Find existing charts first. Show me the proposed charts, definitions, and layout before creating the dashboard.
Expected result: A proposed onboarding dashboard layout using relevant existing charts where possible, with definitions for review.
Use the approved definitions to create the dashboard. Add a short description explaining the audience, date ranges, and interpretation of each chart. Return its Amplitude link.
Expected result: A saved dashboard whose definitions and intended use are clear to the next reader. Review any confirmation cards your client presents.
Analyze an experiment
Provide: the feature experiment URL and the decision you are considering.
Read this experiment, confirm the variants and primary metric, and summarize the available results. Discuss statistical uncertainty and any guardrail trade-offs. Recommend what to investigate next without changing the experiment.
Expected result: A decision brief with the setup, observed results, uncertainty, and suggested follow-up. Changes to feature flags or experiments require the appropriate access.
Review a tracking plan
Provide: the project, tracking-plan branch, and naming or governance rules you want to apply.
Read the tracking plan and find potentially duplicate events, unclear descriptions, and inconsistent property names. Propose a cleanup plan, including affected items and whether each change is reversible.
Expected result: A proposed cleanup plan identifying affected events and properties, governance issues, and the reversibility of each change.
Apply the approved changes on a review branch. Summarize the differences from main and any required confirmations. Leave the branch unmerged for review.
Expected result: A reviewable tracking-plan change set. Read the taxonomy and delete/restore notes before deleting events or properties.
Explore more capabilities with the Developer CLI
Provide: your project, the onboarding gap to address, and any relevant guide or survey identifiers.
Use the supported Developer CLI commands to inspect our existing guides and surveys. Propose an unpublished draft for the onboarding gap we found, and explain the available command limits before making changes.
Expected result: An inventory of existing guides and surveys, a proposed unpublished onboarding draft, and the supported command limits to review before making changes.
For reusable playbooks, browse the open-source skills repository.
Ask your agent to use run_amp to discover and run supported Developer CLI commands for additional Amplitude workflows.
Draft and finish a guide
Provide: your project, the audience and message you want to address, and any existing guide or survey to inspect.
Review the existing guides and surveys for this project, then create an unpublished guide called "Billing refresh announcement" with a single modal step using the content we agreed on. Return the admin URL and explain the remaining audience, preview, and publishing steps in Amplitude.
Expected result: A review of relevant existing items, an unpublished draft on a disabled flag, and its admin URL. Audience configuration, preview, publishing, scheduling, and testing remain in the UI. The agent checks the existing state before retrying any ambiguous write result.
Investigate a failing import
Provide: your project, the source name or type, and the period to investigate.
List the warehouse import sources for this project and resolve my Snowflake connection. Inspect its recent jobs and available configuration. Highlight failed jobs, jobs completed with errors, and successful jobs that rejected rows. Summarize the available failure reasons, counts, timings, and error-log links.
Expected result: Where available, the resolved import source and its configuration, a review of recent jobs across relevant statuses, and available failure evidence. A zero record count is distinguished from an unreported count. Unsupported Salesforce V2 group-property jobs are directed to the source's Ingestion Jobs tab in Amplitude.
Verify a warehouse export
Provide: your project, the destination name or type, and the period to review.
List the export destinations for this project, then inspect the automatic export jobs and the manual or backfill jobs for my Snowflake destination separately. Compare the latest successful job start time with the latest completion time for a run that exported data. Explain whether the available evidence confirms recent delivery.
Expected result: Where available, the resolved export destination, both job modes, and an explanation of recent delivery based on timestamps, reported record counts, and error details. Successful zero-row automatic exports are omitted from job history, so the latest successful job time alone is not treated as proof that rows arrived.
Was this helpful?
For AI agents: a documentation index is available at /docs/llms.txt. Append .md to any page URL for markdown, or send Accept: text/markdown.