Auto-documentation and metric dictionary engine for Tableau, powered by MCP. v1.0
Tableau Scribe connects to your Tableau Cloud or Server environment via the Tableau MCP and generates polished, audience-ready documentation from live views and datasources — no screenshots, no manual data entry, no copy-paste.
Tableau Scribe operates in two modes:
Point it at any published Tableau view and receive structured dashboard documentation covering purpose, key metrics, filters, usage guidance, and a lightweight design health summary. The skill fetches the rendered image, view metadata, workbook context, and datasource schema automatically.
Point it at any published datasource and receive a metric dictionary with plain-language definitions, directionality, business context, and field classifications. Designed so business users can understand what each metric means without opening Tableau.
- Two depth levels — Brief (1-page reference card) or Detailed (full 3-page documentation)
- Three output formats —
.docx(Word), on-screen (inline markdown), or.md(downloadable file) - Zero-upload workflow — No screenshots needed; MCP pulls the view image and metadata directly
- Graceful degradation — If an API call fails (401, timeout, permission issue), Scribe still produces what it can and explains gaps in a bottom note rather than aborting
- Inline detection — If you specify format or depth in your request ("give me a brief Word doc"), Scribe skips the questions and goes straight to work
- Stale metadata detection — Flags when datasource schema and rendered view are out of sync
- Execution time transparency — Tells you upfront how long to expect (~15s brief, ~30-60s detailed)
- Claude Desktop with agent mode enabled (Desktop
- Tableau MCP configured and authenticated against your Tableau Cloud or Server site
- Permissions — Your Tableau user needs at least Viewer access to the views/datasources you want to document
Trigger Tableau Scribe with natural language:
Document this dashboard: [view name or URL]
Generate a metric dictionary for [datasource name]
/tableau-scribe
| Prompt | Mode | What Happens |
|---|---|---|
| "Document the Sales Overview dashboard as a Word doc" | Auto-Doc | Detects .docx format, asks depth, generates documentation |
| "Give me a brief metric dictionary for the Segment Consumption datasource" | Metric Builder | Detects brief depth, asks format, generates dictionary |
| "Write detailed documentation for this view on screen" | Auto-Doc | Skips both questions (format + depth detected), proceeds immediately |
| "Tableau scribe — document everything about my Data Cloud dashboard" | Auto-Doc | Detects detailed depth, asks format |
A single-page quick reference: purpose, top metrics with directionality, filters, how-to-read guidance, and refresh notes.
Seven sections: Purpose & Audience, Key Metrics (with definitions, business meaning, and directionality), Filters & Interactivity, How to Use This Dashboard, Data & Refresh, Design Health Summary, and Glossary.
Table-based one-liners: measures, calculated fields, and parameters — one row each with definition and direction.
Full field-level documentation: primary measures with business context and typical ranges, calculated fields with plain-language logic explanations, dimensions with cardinality, date fields with granularity, parameters with allowable values, and field relationship notes.
Tableau Scribe is designed to work alongside other Tableau MCP skills:
Blueprint ──→ Build ──→ Scribe ──→ VizCritique
(design) (create) (document) (evaluate)
| Skill | Relationship |
|---|---|
| Tableau Dashboard Blueprint | Designs the dashboard → Scribe documents what was built |
| Tableau Calc Master | Writes calculations → Scribe explains them in plain language |
| Tableau Auto-Query | Provides sample data → Scribe uses it to enrich metric definitions |
| VizCritique Pro | Full scored evaluation → Scribe includes only a lightweight health summary and defers to VizCritique for depth |
Scribe never aborts because one API call failed. Instead:
- It produces documentation from whatever data is available (graceful degradation across 5 priority levels)
- It places a bottom note at the end of the document explaining what was unavailable and why — keeping the documentation body clean
- It marks individual fields with "(inferred)" or "(not currently visible)" when specific items can't be confirmed
This means you always get output, even when your Tableau session has expired or a datasource is permission-restricted.
| Version | Changes |
|---|---|
| v1.3 | Depth question previews, execution time expectations, stale metadata detection |
| v1.2 | Brief/Detailed depth modes, bottom-note error pattern, format question upfront, inline detection |
| v1.1 | Structural enhancements from independent scoring |
| v1.0 | Initial release — dual-mode Auto-Doc + Metric Builder |
Adam Mico — Built on the documentation methodology from Dashboard Documentation Pro, adapted for the Tableau MCP ecosystem.
This skill is provided as-is for use within the Claude Desktop agent environment. See your organization's Tableau MCP licensing for API access terms.