Imported from amplitude/mcp-marketplace (plugins/amplitude/skills/chart-link-analysis/SKILL.md). Install upstream with npx skills add amplitude/mcp-marketplace --skill chart-link-analysis. Copyright stays with the author.

Chart Link Analysis

Second most common job: the user pastes a link and wants the analysis understood, refreshed, or adapted.

The arc

1. Resolve the URL. get_from_url returns the entity type, IDs, and (for charts) the current definition. One call handles chart, dashboard, experiment, notebook, and cohort links — use it instead of asking the user for IDs.
2. Read the definition. For a chart: get_amplitude_charts with include: 'typed' — the typed params mirror the chart-builder UI and are the fastest way to see how the project spells events/properties. If typed fails to lower the chart ("advanced features are not modeled"), fall back to include: 'definition' for the raw form. For a dashboard: use_amp_dashboards action: 'get' (batch ≤3) to get its chart list, then read the charts that matter.
3. Pull the data. get_amplitude_charts include: 'data' with the chart IDs, ≤3 IDs per call. For a report refresh, request the granularity the report needs (daily for week-over-week, weekly for monthlies).
4. Adapt or fork. Edit the typed params, then call query_amplitude_data with the edited chart and chartId set to the saved chart's id — the parent's params fill gaps the typed model omits, and the edit stays linked to its parent. Render the result with render_amplitude_chart; the user saves it from the widget's Save chart button (there is no model-facing save tool). To replicate a whole dashboard in a new context: read the reference dashboard, then rebuild each chart — do not try to "copy" in one call.

Expectations to set

• get_from_url does not run the chart — it returns metadata/definition. Data always comes from get_amplitude_charts include: 'data' (saved charts) or query_amplitude_data (ad hoc).
• include: 'data' returns the same numbers the UI shows, including Amplitude's timezone bucketing — when comparing against external sources (e.g. BigQuery), call out timezone and partial-day differences explicitly.
• Definition reads are cheap; data reads are the expensive calls. Read all definitions first, then decide which 1–3 charts actually need data.
• Just need the link for the user? include: 'link' (the default) validates the ids and returns URLs without running anything.