amplitude-navigator

name	amplitude-navigator
description	Comprehensive guide for Amplitude MCP queries. Use when analyzing product analytics, user behavior, funnels, retention, experiments, or metrics. Reduces context overhead by documenting query patterns, event taxonomies, and RDC-specific configurations. Triggers on Amplitude, product analytics, user behavior, funnel analysis, retention, experiment analysis, or metric queries.

Amplitude Navigator

This skill provides efficient navigation of the Amplitude MCP tools, reducing context overhead by documenting common query patterns, event taxonomies, and RDC-specific configurations.

RDC Context

Organization

• Organization ID: 228133
• Organization URL: realtor
• Organization Name: Realtor.com

Key Projects

Project	App ID	Description	Use Case
Realtor - Production	558383	Primary clickstream data	Main analytics, experiments
Realtor - Leads 2.0	678364	Lead submission data	Lead analysis, attribution, click attributed EFR
Consumer Marketing Data Braze	674963	Braze notification events	Push/email engagement
Consumer Marketing Data Cordial	678109	Cordial email events	Email performance
Real Time SDK - Prod	675822	Browser SDK with Session Replay	Session replay, real-time
Realtor - QA	645393	QA/Alpha/Beta environments	Testing only
Seller Leads Exploration	717063	Seller vertical data	Seller analytics

Default Project: 558383 (Realtor - Production)

Tool Selection Matrix

Task	Tool	When to Use
Discover content	search	Find existing charts, events, metrics, experiments, dashboards
Ad-hoc analysis	query_dataset	Custom queries not covered by existing charts
Existing chart data	query_charts	When you have chart IDs (max 3 per call)
Saved metric data	query_metric	Query predefined metrics by ID
Experiment results	query_experiment	A/B test statistical analysis
Event properties	get_event_properties	Get properties for a specific event type
Full experiment details	get_experiments	Retrieve experiment configuration, state, decisions
Dashboard contents	get_dashboard	Get all charts in a dashboard
Session recordings	get_session_replays	Find user session recordings
URL lookup	get_from_url	Parse Amplitude URLs to get objects

Workflow Pattern

1. Search first - Use search to discover existing charts/events before building custom queries
2. Get properties - Use get_event_properties after finding exact event names
3. Query or build - Use existing charts (query_charts) or build custom (query_dataset)
4. Always reference - Include chart/metric links in responses for validation

Query Types Reference

Events Segmentation (Trends)

For time-series analysis of events.

{
  "name": "Daily Active Users",
  "type": "eventsSegmentation",
  "app": "558383",
  "params": {
    "range": "Last 30 Days",
    "events": [{"event_type": "_active", "filters": [], "group_by": []}],
    "metric": "uniques",
    "countGroup": "User",
    "groupBy": [],
    "interval": 1,
    "segments": [{"conditions": []}]
  }
}

Metric Options:

• uniques - Unique users
• totals - Event count
• average - Per user average
• pct_dau - % of DAU
• frequency - Distribution
• sums - Property sum (requires property in group_by)
• value_avg - Property average

Funnels (Conversion Analysis)

For step-by-step conversion analysis.

{
  "name": "Lead Funnel",
  "type": "funnels",
  "app": "558383",
  "params": {
    "range": "Last 30 Days",
    "events": [
      {"event_type": "pageview", "filters": [], "group_by": []},
      {"event_type": "cobrokelead", "filters": [], "group_by": []}
    ],
    "countGroup": "User",
    "segments": [{"conditions": []}]
  }
}

Key Parameters:

• conversionWindow: {"value": 1, "unit": "day"} - Time limit for completion
• order: "this_order" (default), "any_order", "exact_order"

Retention

For cohort retention analysis.

{
  "name": "New User Retention",
  "type": "retention",
  "app": "558383",
  "params": {
    "range": "Last 90 Days",
    "startEvent": {"event_type": "_new", "filters": [], "group_by": []},
    "retentionEvents": [{"event_type": "_active", "filters": [], "group_by": []}],
    "retentionMethod": "nday",
    "countGroup": "User",
    "interval": 1,
    "segments": [{"conditions": []}]
  }
}

Retention Methods:

• nday - Return on specific day
• rolling - Return on or after day
• bracket - Custom ranges via retentionBrackets

Sessions

For session-based metrics.

{
  "name": "Average Session Length",
  "type": "sessions",
  "app": "558383",
  "params": {
    "range": "Last 30 Days",
    "sessions": [{"filters": [], "group_by": []}],
    "countGroup": "User",
    "sessionType": "average",
    "segments": [{"conditions": []}]
  }
}

Session Types: average, totalSessions, peruser, averageTimePerUser, totalTime, length

Meta Event Types

Special system events available in all queries:

Event Type	Description	Use Case
_active	Any active event	DAU, MAU, active users
_new	First event by new users	New user analysis, retention start
_all	Any tracked event	Total event volume
_any_revenue_event	Revenue events	Revenue analysis
$popularEvents	Top events by volume	Taxonomy exploration

Property Syntax

Property Sources

Source	Prefix	Example
Amplitude Core	None	country, platform, device_type
Customer (Custom)	gp:	gp:vertical, gp:page_type
Experiment Flags	gp:[Experiment]	gp:[Experiment] MBL2510_FEATURE

Common Amplitude Core Properties

User/Session: user_id, amplitude_id, device_id, session_id Geographic: country, city, region, dma Device: platform, device, device_type, os, language Application: version, start_version, library

Filter Syntax

Segment-level filter (applies to all events):

"segments": [{
  "conditions": [{
    "type": "property",
    "group_type": "User",
    "prop_type": "user",
    "prop": "country",
    "op": "is",
    "values": ["United States"]
  }]
}]

Event-level filter (applies to specific event):

"events": [{
  "event_type": "pageview",
  "filters": [{
    "group_type": "User",
    "subprop_key": "page_type",
    "subprop_op": "is",
    "subprop_type": "event",
    "subprop_value": ["ldp"]
  }],
  "group_by": []
}]

Filter Operators: is, is not, contains, does not contain, set is, set is not, greater than, less than

Group By Syntax

Event property:

"group_by": [{"type": "event", "value": "platform"}]

User property:

"group_by": [{"type": "user", "value": "country"}]

Top-level groupBy (across all events):

"groupBy": [{"type": "user", "value": "country", "group_type": "User"}]

Time Parameters

Range Strings

• "Last N Days" - Last 7 Days, Last 30 Days, Last 90 Days
• "Last N Weeks" - Last 4 Weeks
• "Last N Months" - Last 3 Months
• "This Week", "This Month", "This Quarter", "This Year"

Unix Timestamps

For precise boundaries:

"start": 1727740800,
"end": 1730419199

ISO Dates

"start": "2025-01-01",
"end": "2025-01-31"

Interval Values

Value	Granularity
-3600000	Hourly
1	Daily (default)
7	Weekly
30	Monthly
90	Quarterly

Important: Only these exact values are valid. Arbitrary intervals (2, 3, 14) will error.

Common RDC Events

Lead Events

Event	Description
cobrokelead	Cobroke listing lead submission
advantagelead	Advantage listing lead
rentallead	Rental property lead
farlead	Find a Realtor lead
newconstructiondirectlead	New construction lead
forsaleagentconnection	Agent connection lead

Engagement Events

Event	Description
pageview	Page view
search	Search performed
refinedsearch	Search filter applied
click	Generic click
saveditem	Listing/search saved
listingclick	Listing card click

User Events

Event	Description
signup	Account created
signin	User signed in
signout	User signed out
claimhome	Home claimed

Experiment Events

Event	Description
[Experiment] Exposure	Experiment variant exposure
[Experiment] Assignment	Variant assignment
expexposure	Legacy exposure event

Experiment Analysis

Querying Experiments

1. Search for experiment:

Amplitude:search(entityTypes=["EXPERIMENT"], query="feature_name")

2. Get experiment details:

Amplitude:get_experiments(ids=["123456"])

3. Query results (primary metric only by default):

Amplitude:query_experiment(id="123456")

4. With segment breakdown:

Amplitude:query_experiment(
  id="123456",
  groupBy=[{"type": "user", "value": "platform", "group_type": "User"}]
)

Interpreting Results

Key Fields:

• pValue - Statistical significance (< 0.05 = significant)
• relativeLift - Percentage change vs control
• absoluteLift - Raw metric difference
• ssrmTimeseries.srmDetected - Sample Ratio Mismatch flag
• recommendation - "rollout", "rollback", or null

Decision Framework:

• Significant + Positive Lift + No SRM → Rollout
• SRM Detected → Rollback (data quality issue)
• >30 days + p=1.0 → Consider terminating
• <14 days → Continue running

Session Replay

Basic Search

Amplitude:get_session_replays(
  projectId="558383",
  segmentFilters=[{"conditions": []}],
  limit=10
)

By Experiment Variant

Amplitude:get_session_replays(
  projectId="558383",
  segmentFilters=[{
    "conditions": [{
      "type": "property",
      "group_type": "User",
      "prop_type": "user",
      "prop": "gp:[Experiment] feature_flag",
      "op": "is",
      "values": ["treatment"]
    }]
  }],
  limit=10
)

With Event Filters

Amplitude:get_session_replays(
  projectId="558383",
  eventCountFilters=[{
    "count": "1",
    "operator": "greater or equal",
    "event": {"event_type": "cobrokelead", "filters": [], "group_by": []}
  }],
  limit=10
)

Response Parsing

CSV Response (isCsvResponse: true)

• Header rows contain metadata
• Data header row has column labels (dates)
• Data rows: label columns + value columns
• Values may be prefixed with \t

JSON Response (isCsvResponse: false)

{
  timeSeries: [[{value: 614}, {value: 1769}]],
  overallSeries: [[{value: 2383}]],
  seriesMetadata: [{segmentIndex: 0, formulaIndex: 0}],
  xValuesForTimeSeries: ["2025-01-01T00:00:00", "2025-01-02T00:00:00"]
}

Reference Documents

• query_patterns.md - Detailed query templates
• events_taxonomy.md - Complete event reference

Tips for Efficient Queries

1. Search before building - Existing charts are pre-optimized
2. Always name queries - Include descriptive name field
3. Use correct project - Production (558383) vs Leads (678364)
4. Limit group by values - Use groupByLimit to control cardinality
5. Exclude incomplete data - Set excludeIncompleteDatapoints: true for accurate trends
6. Reference charts in responses - Always provide links for validation
7. Batch experiment queries - Query 3-5 at a time to avoid timeouts