svelte-component-graph-mcp
An MCP server that exposes a Svelte/SvelteKit project's component dependency graph over stdio, enabling AI coding assistants to query component imports, unused components, and route dependencies without an editor.
README
svelte-component-graph-mcp
An MCP server that exposes a Svelte/SvelteKit project's
component dependency graph over stdio, so an AI coding assistant can ask questions like
"what imports Button.svelte?", "which components are unused?", or "what does this route pull
in?" without opening an editor.
It is the command-line companion to the
Svelte Component Visualizer
VS Code extension and uses the same analysis: the Svelte compiler's AST + estree-walker to track
default and named .svelte imports, resolve static <svelte:component this={Identifier}> usage,
and flag imported-but-unused components.
What it gives you
A graph of every component and route in the project:
- Nodes are keyed by workspace-relative path (
src/lib/Button.svelte) — so two files that share a name are never conflated — with a human-readablelabel, atype(component|route), and anunusedflag. - Links are import edges (
sourceimportstarget).
The server is stateless across projects: every tool takes a root argument, so one running
server can answer questions about many projects (and many git worktrees) at once. Results are cached
per root and refreshed automatically when files change (checked on each call by mtime + size; only
changed files are re-parsed).
Installation
This is a stdio MCP server — your MCP client launches it; you don't run it by hand. Requires Node.js ≥ 20. Pick the path that matches your setup.
Claude Code — plugin (recommended)
Installs the MCP server and the companion /svelte-graph skill in one step:
/plugin marketplace add jamcgrath/svelte-component-graph-mcp
/plugin install svelte-component-graph@jamcgrath
Restart Claude Code afterward so the server connects. Nothing to clone or build — it runs via npx.
(The two commands can be combined: /plugin install svelte-component-graph@jamcgrath/svelte-component-graph-mcp.)
Claude Code — server only (no skill)
claude mcp add svelte-graph --scope user -- npx -y svelte-component-graph-mcp
Other MCP clients (Claude Desktop, Cursor, Cline, …)
Add it to the client's MCP config:
{
"mcpServers": {
"svelte-graph": {
"command": "npx",
"args": ["-y", "svelte-component-graph-mcp"]
}
}
}
(If you npm install -g svelte-component-graph-mcp, use "command": "svelte-component-graph-mcp" with
"args": [].)
Quick test, no client
Drive it with the MCP Inspector:
npx @modelcontextprotocol/inspector npx -y svelte-component-graph-mcp
The root argument
Every tool requires root: an absolute path to the project you want analyzed. The server
validates that it exists, is a directory, and contains at least one .svelte file.
Guidance for the assistant when choosing root:
- If a
svelte.config.jsexists in the current working directory, the project root is the cwd — pass$PWD. - Otherwise (e.g. a monorepo, or the cwd is a subfolder), pass the project's absolute path explicitly.
Companion skill
A Claude Code skill (skills/svelte-graph/) teaches the assistant
when to reach for these tools (impact analysis, dead-code hunts, prop lookups) and how to resolve
root. The MCP tools work without it; the skill just makes the assistant use them more readily.
If you installed the plugin, you already have it — nothing to do. Otherwise, copy the folder into a skills directory:
# user-level (all your projects)
cp -r skills/svelte-graph ~/.claude/skills/svelte-graph
# or project-level (this repo only)
cp -r skills/svelte-graph .claude/skills/svelte-graph
Then invoke it with /svelte-graph, or let the assistant trigger it automatically.
Tools
get_graph(root)
The full graph.
{
"nodes": [
{ "id": "src/routes/+page.svelte", "label": "(page) /", "type": "route" },
{ "id": "src/lib/Button.svelte", "label": "Button", "type": "component" },
{ "id": "src/lib/Unused.svelte", "label": "Unused", "type": "component", "unused": true }
],
"links": [
{ "source": "src/routes/+page.svelte", "target": "src/lib/Button.svelte" }
]
}
get_component(root, path)
Details for one component/route, including its public API surface. path is workspace-relative
(src/lib/Button.svelte).
{
"id": "src/lib/Widget.svelte",
"label": "Widget",
"type": "component",
"unused": false,
"isRoute": false,
"parents": ["src/routes/+page.svelte"], // components that import it
"children": [], // components it imports
"props": [ // from $props()
{ "name": "size", "optional": true, "bindable": false },
{ "name": "open", "optional": true, "bindable": true },
{ "name": "title", "optional": false, "bindable": false },
{ "name": "rest", "optional": true, "bindable": false, "rest": true }
],
"slots": ["default", "footer"] // <slot> / <slot name="…">
}
Props come from Svelte 5 runes ($props(), $bindable(), ...rest). Events are not a separate
concept in Svelte 5 — they are ordinary callback props, so they appear in props.
get_unused(root)
Every component imported somewhere but never used in the importing file's template.
[
{ "id": "src/lib/Unused.svelte", "label": "Unused", "type": "component", "unused": true }
]
get_routes(root)
Every route (+page / +layout / +error) with the components it directly pulls in.
[
{
"id": "src/routes/dashboard/+page.svelte",
"label": "(page) /dashboard",
"children": ["src/lib/Icon.svelte", "src/lib/components/Button.svelte"]
}
]
scan(root)
Force a full re-parse (bypassing all caches) and return the resulting size.
{ "nodes": 7, "links": 5 }
How resolution works (and its limits)
- Relative imports (
./,../) resolve against the importing file's directory. - SvelteKit's
$lib/…resolves to the nearestsrc/lib(monorepo-aware — it uses the importing file's ownsrc/, not a global root). - Other bare/aliased specifiers (custom Vite aliases beyond
$lib) are kept as best-effort leaf nodes rather than resolved to a file. As a corollary, a bare package import likeimport X from 'some-pkg/Foo.svelte'is keyed by that raw path and would merge with a local file at the same relative path (some-pkg/Foo.svelte) if one exists — an unlikely but possible collision. <svelte:component this={…}>is resolved only whenthisis a plain imported identifier; dynamic expressions (member access, conditionals) are not traced.componentPaths/routePathsaccept negation globs (!**/*.stories.svelte) to exclude files: a file is included when it matches a positive pattern and no negation pattern.propsare read from the$props()object destructuring. A non-destructured binding (let props = $props()) has no statically-known prop names, sopropscomes back empty for it.
License
MIT © James McGrath
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.