vue-harvest-mcp
vue-harvest-mcp
MCP server for AI-assisted Vue component extraction and design system analysis.
This is the intelligence layer for vue-harvest. It picks up where the CLI leaves off, handling the components in the 30-70% confidence range that need reasoning to extract safely.
Setup
Claude Desktop
Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"vue-harvest": {
"command": "vue-harvest-mcp"
}
}
}
Claude Code
claude mcp add vue-harvest vue-harvest-mcp
Cursor / other MCP clients
npx vue-harvest-mcp
The server communicates over stdio using the Model Context Protocol.
Tools
The server exposes 12 tools:
Analysis
| Tool | Description |
|---|---|
analyze-project |
Runs the full vue-harvest pipeline on a project. Required before using other tools. Takes a path argument. |
list-components |
Lists components with optional filters: tier, minConfidence, maxConfidence. |
inspect-component |
Returns the full analysis of a component including source code, interface, dependencies, and coupling issues. |
coupling-report |
Generates a project-wide coupling report grouped by issue type. Good for spotting systemic patterns. |
Extraction
| Tool | Description |
|---|---|
extract-component |
Extracts a component into a standalone bundle. Set force: true to extract below threshold. |
deep-analyze |
Structured coupling analysis with source code. Returns data the LLM can reason about to suggest fixes. |
suggest-refactor |
Generates concrete refactoring suggestions with before/after code snippets. |
generate-wrapper |
Creates a composable wrapper that abstracts store access away from a component. |
adapt-and-extract |
Rewrites a component to remove all coupling, then extracts it. Best for the 30-70% confidence range. |
batch-triage |
Triages all reviewable components at once with prioritized effort estimates. |
Design system
| Tool | Description |
|---|---|
analyze-design-system |
Extracts design tokens (colors, typography, spacing, shadows) from a project. |
get-design-tokens |
Returns extracted tokens, optionally filtered by type (color, font-size, spacing, etc). |
Resources
After running analyze-project, these resources become available:
harvest://registry: full component registry JSONharvest://graph: dependency graph (nodes, edges, cycles)harvest://summary: analysis summaryharvest://component/{name}: individual component analysisharvest://design-system: extracted design tokens
Prompts
| Prompt | Description |
|---|---|
analyze-new-project |
Guided first analysis of a Vue project |
extraction-sprint |
Batch refactor and extract session |
refactor-component |
Single component deep refactor |
extract-design-system |
Design token extraction and analysis |
Typical workflow
- Ask the LLM to analyze your project. It will run
analyze-project. - Ask what needs review. It will use
list-componentsandbatch-triage. - For each reviewable component, the LLM can
deep-analyzeit, figure out the right approach, and eithersuggest-refactor,generate-wrapper, oradapt-and-extract. - For design tokens, ask the LLM to run
analyze-design-systemand recommend a naming convention.
Example prompts you can use
"Analyze my Vue project at /path/to/project"
"Show me all components between 40% and 70% confidence"
"Deep analyze the UserProfileCard and suggest how to remove the store dependency"
"Do an extraction sprint, start with the easiest wins"
"Extract the design tokens and suggest a consistent naming scheme"
Requirements
- Node.js 20+
- An MCP-compatible client (Claude Desktop, Claude Code, Cursor, etc)
License
MIT
Related Packages
MCP server for building Vue.js design systems - component analysis, documentatio...
Compile a Vue SFC into a self-contained ui:// HTML resource for an MCP App.
Vue renderer for MCP Apps: useToolResult/useCallTool/useTheme hooks + renderer a...
Version History
| Version | Released | License |
|---|---|---|
| 0.0.4 | MIT | |
| 0.0.3 | MIT | |
| 0.0.2 | MIT | |
| 0.0.1 | MIT |