Loading ToolBench
MCP server for growth marketing — campaign design, retention analysis, churn prediction
Growth MCP provides 9 well-structured tools focused on growth marketing analytics. All tools have descriptive text and explicit input schemas with typed parameters. However, descriptions lack strategic guidance on WHEN to use each tool vs alternatives, parameter descriptions are minimal (mostly 1-2 line specs without format/constraint details), and output schemas are not documented. No tool annotations (readOnlyHint, destructiveHint) despite all being read-only. Error handling returns JSON with error messages but lacks recovery guidance. Overall solid foundation but misses LLM-optimization opportunities.
Run an A/B test analysis on the result of a BigQuery SELECT query. Requires: pip install "growth-mcp[bigquery]" and Google ADC auth. The query must return 1 row per user with a group column and a conversion flag column. Read-only: only SELECT/WITH queries allowed.
Run an A/B test analysis directly from a raw CSV (one row per user). Aggregates conversions per group and runs a two-proportion z-test. Use analyze_experiment instead if you already have aggregated counts.
Run an A/B test analysis straight from Mixpanel events. Requires MIXPANEL_API_SECRET env var (no extra pip dependency). Exposure events assign users to groups via group_property; conversion events mark who converted. Date range capped at 90 days.
Run retention cohort analysis on a BigQuery query result. Requires: pip install "growth-mcp[bigquery]" and Google ADC auth. The query must return 1 row per period with a period label and a retention rate (0-1) or active user count (auto-normalized).
Run retention cohort analysis from a CSV of periods and rates or counts. Accepts retention rates (0.0-1.0) or active user counts per period (counts are auto-normalized against the first period).
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract results with confidence.
Parameter descriptions are minimal (1-2 lines) and lack format/constraint guidance. E.g., 'Column holding the period label' doesn't explain valid period_col naming conventions, regex patterns, or case sensitivity.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | 2026-07-28+ | v2 |
Inspect a local CSV file: columns, inferred types, row count, sample rows. Run this first before any *_from_csv tool to discover column names.
Per-segment statistics on a BigQuery query result. Requires: pip install "growth-mcp[bigquery]" and Google ADC auth. Useful for segmented balance or spend analysis on warehouse data.
Per-segment statistics (count, sum, mean, min, max, share) from raw CSV rows. Useful for segmented balance or spend analysis, e.g. loyalty point balances or voucher redemption value by user segment.
Per-segment statistics over Mixpanel events. Requires MIXPANEL_API_SECRET env var. Useful for revenue or redemption value by segment. Date range capped at 90 days.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being read-only. This is critical metadata for agent planning and safety.
Missing WHEN guidance in tool descriptions. Descriptions explain WHAT but not when to select one tool vs another similar one. E.g., when to use analyze_experiment_from_csv vs analyze_experiment_from_bigquery?
Error handling returns structured JSON error objects but provides no recovery guidance. E.g., 'file not found' doesn't suggest inspecting CSV first or checking file path format.
BigQuery and Mixpanel tools require external auth/credentials (ADC, MIXPANEL_API_SECRET) but descriptions don't explain authentication troubleshooting or error recovery paths.