Holoviz Mcp Cli

holoviz-mcp CLI

The holoviz-mcp CLI provides direct access to HoloViz documentation, component introspection, and visualization tools from the command line.

Installation

The recommended way to install is as a uv tool:

uv tool install holoviz-mcp

This installs both holoviz-mcp and hv commands into an isolated environment. Use hv as a shorter alternative in all examples below (e.g., hv search Panel Tabulator, hv pn get Button).

When to Use CLI vs MCP

• Prefer MCP tools when the HoloViz MCP server is loaded — they return structured, typed data (including images) and integrate natively with LLM tool calling
• Use CLI when MCP tools are not available but holoviz-mcp (or hv) is installed and you have Bash/shell access
• Search the web as a last resort when neither MCP tools nor CLI are available

Output Formats

All tool commands support three output formats via --output / -o:

• pretty (default) — Rich terminal output with colors and tables, for humans
• markdown — Markdown-formatted text, optimized for LLMs
• json — Structured JSON, for scripts and programmatic use

Setup commands (serve, install, update) output human-readable text by default.

Command Structure

Commands follow noun-verb pattern with Python import aliases for library-specific tools:

holoviz-mcp <command> # general tools holoviz-mcp pn <command> # Panel tools (import panel as pn) holoviz-mcp hv <command> # HoloViews tools (import holoviews as hv) holoviz-mcp hvplot <command> # hvPlot tools (import hvplot)

Quick Reference

Search & Documentation

holoviz-mcp search Panel Tabulator # search all indexed docs (no quotes needed) holoviz-mcp search query -p panel # search within a project holoviz-mcp search query -n 5 # return up to 5 results holoviz-mcp doc get <path> <project> # get specific document (use search to find paths) holoviz-mcp ref get <component> # find reference guide (when you know the exact name) holoviz-mcp ref get <component> -p panel # filter by project

Skills

holoviz-mcp skill list # list available skills with descriptions holoviz-mcp skill get <name> # get skill content (always Markdown)

Panel (pn)

holoviz-mcp pn list # list all components holoviz-mcp pn list -P panel_material_ui # filter by package holoviz-mcp pn list -m panel.widgets # filter by module path holoviz-mcp pn get Button -P panel # get full component details holoviz-mcp pn params Button -P panel # get parameter details holoviz-mcp pn search input widget # search components by keyword (no quotes needed) holoviz-mcp pn search chart -l 5 # limit results holoviz-mcp pn packages # list Panel-related packages

hvPlot

holoviz-mcp hvplot list # list plot types (~28 types) holoviz-mcp hvplot get scatter # get plot-specific params (compact) holoviz-mcp hvplot get scatter --generic # include generic options shared by all plot types holoviz-mcp hvplot get scatter --style bokeh # include backend-specific style options holoviz-mcp hvplot get scatter --signature # get function signature instead

HoloViews (hv)

holoviz-mcp hv list # list visualization elements (~60 elements) holoviz-mcp hv get Curve # get Curve element docstring holoviz-mcp hv get Curve --backend matplotlib # specify backend

Projects

holoviz-mcp project list # list indexed projects (~20 projects)

Display & Inspect

holoviz-mcp inspect # inspect app at localhost:5006 holoviz-mcp inspect http://localhost:8080 # inspect specific URL holoviz-mcp inspect --no-screenshot # console logs only holoviz-mcp inspect --no-console-logs # screenshot only holoviz-mcp inspect --full-page --delay 5 # full page with 5s delay

Index Management

holoviz-mcp update index # refresh documentation index

Tips

• Use holoviz-mcp search with specific terms for better results: "Tabulator pagination page_size" > "how to paginate a table"
• Use holoviz-mcp pn get when you need the full docstring, constructor signature, and parameters
• Use holoviz-mcp pn params when you only need parameter details (lighter, no docstring)
• Use holoviz-mcp ref get when you know the exact component name; use search for fuzzy/semantic lookup
• Pipe JSON output to jq for filtering: holoviz-mcp pn list -o json | jq '.[].name'
• Use -o pretty for human-readable output in the terminal