unity-asset-reference-mcp
Indexes Unity project assets into a SQLite reference graph and provides an MCP server for querying asset dependencies, references, unused assets, and broken references.
README
unity-asset-reference-mcp
Index a Unity project's assets into a reusable SQLite reference graph, then query it three ways: a CLI, an MCP server (for Claude and other agents), and a local web viewer. Unity-only.
- Impact analysis — "what references this material?" before you change it
- Unused-asset detection — orphans nothing loads (Addressables-aware)
- Dependency tracing — the full closure a scene/prefab pulls in
- Broken-reference detection — refs to deleted/missing assets
- Reusable artifact — a plain
.sqliteany tool can open; no lock-in
It reads Unity's own serialization (.meta GUIDs + YAML {fileID, guid} refs),
so the graph is exact — no heuristics, no AST. Requires Asset Serialization:
Force Text (Unity's default for version control).
Requirements
- Node.js ≥ 20
- A Unity project using Force Text serialization
better-sqlite3builds via prebuilt binaries during install (native module)
Install
npm install -g unity-asset-reference-mcp
This puts three commands on your PATH:
| Command | What |
|---|---|
unity-asset-reference-mcp-index |
the CLI indexer |
unity-asset-reference-mcp |
the MCP server |
unity-asset-reference-mcp-web |
the web viewer |
Without a global install (npx): the MCP server's bin name matches the package,
so npx -y unity-asset-reference-mcp … works directly. For the other two bins,
name the package with -p:
npx -y unity-asset-reference-mcp --project /path/to/UnityProject # MCP server
npx -y -p unity-asset-reference-mcp unity-asset-reference-mcp-index index /path # indexer
npx -y -p unity-asset-reference-mcp unity-asset-reference-mcp-web --db <index.db> # viewer
1. Index a project
Builds <project>/.asset-memory/index.db.
unity-asset-reference-mcp-index index /path/to/UnityProject --force
Add this to your Unity project's .gitignore — ignore the live index, but
commit the config and (optional) shared snapshot:
# unity-asset-reference-mcp: ignore the live index, keep config + shared snapshot
.asset-memory/index.db
.asset-memory/index.db-*
.asset-memory/*.building-*
2. MCP server (works with any MCP client)
This is a standard stdio MCP server — it works with any MCP-compatible host: Claude Code/Desktop, Cursor, Windsurf, Cline, VS Code (Copilot agent), Zed, and others. Nothing is Claude-specific; only where you put the config differs.
Generic config (Claude Desktop, Cursor, Windsurf, Cline, and most hosts use
this mcpServers shape):
{
"mcpServers": {
"unity-asset-graph": {
"command": "npx",
"args": ["-y", "unity-asset-reference-mcp", "--project", "/path/to/UnityProject"]
}
}
}
Where that config lives, per host:
| Host | Config location |
|---|---|
| Claude Code | claude mcp add unity-asset-graph -- npx -y unity-asset-reference-mcp --project /path/to/UnityProject (or .mcp.json in the project) |
| Claude Desktop | claude_desktop_config.json |
| Cursor | .cursor/mcp.json (project) or ~/.cursor/mcp.json (global) |
| Windsurf | ~/.codeium/windsurf/mcp_config.json |
| Cline | cline_mcp_settings.json |
| VS Code (Copilot) | .vscode/mcp.json — uses the key servers instead of mcpServers |
Prerequisites: Node ≥ 20 on PATH (for npx), and a Unity project on Force
Text serialization. You do not need to pre-index — call the index_project
tool once from the agent and it builds <project>/.asset-memory/index.db; the
read tools return a clear no-index error until you do.
Tools exposed: index_project, index_status, get_dependencies,
find_references, find_unused_assets, trace_path, search_assets,
get_overview.
3. Web viewer
# server flavor — serves the viewer + a JSON API over the index
unity-asset-reference-mcp-web --db /path/to/UnityProject/.asset-memory/index.db
# open http://localhost:7777
There is also a static flavor: open dist/web/public/viewer.html in a
browser and pick a .db — it runs the same queries entirely in-browser (WASM
SQLite), no server. Both share one query layer.
Team sharing (snapshots)
Instead of every teammate re-indexing from scratch, commit a compressed
snapshot and let them restore it — the same idea as codebase-memory's
.codebase-memory/.
# after indexing, export a shareable snapshot (or pass --snapshot to `index`)
unity-asset-reference-mcp-index snapshot /path/to/UnityProject
# -> .asset-memory/index.db.br (brotli, ~85% smaller) + artifact.json + .gitattributes
# a teammate who clones + pulls restores the live index with no re-index:
unity-asset-reference-mcp-index restore /path/to/UnityProject
The MCP server and web viewer auto-restore from a snapshot on first use if the live index is missing, so a fresh clone "just works". Commit these:
.asset-memory/config.json # per-project settings (below)
.asset-memory/index.db.br # compressed shared index
.asset-memory/artifact.json # snapshot metadata (schema, commit, counts)
.asset-memory/.gitattributes # marks the blob binary + merge=ours
artifact.json records the git commit and counts the snapshot was built at, so
you can tell when it's stale and re-run index --snapshot.
Configuration
Optional per-project .asset-memory/config.json (see
docs/product/configuration.md):
{
"unused": { "addressableRoots": "auto" },
"scan": { "ignore": ["**/ThirdParty/**", "*.bak"], "ignoreDefaults": true }
}
unused.addressableRoots(auto|on|off) — count Addressable entries as roots for unused detection (query-time;auto= on if the project uses Addressables). Overridable per call.scan.ignore/scan.ignoreDefaults— extra ignore globs (index-time).
The SQLite artifact
Open .asset-memory/index.db with any SQLite tool. Core tables: assets
(nodes), edges (references), unresolved_refs (broken refs),
addressable_entries, index_meta. Schema:
docs/product/asset-graph-model.md.
Known limitations
- Code-based refs not tracked:
Resources.Load("path")and hard-coded Addressable address strings in C# aren't scanned yet, sofind_unusedoutput is candidates — verify against your loading code. - Asset-level granularity: edges are asset→asset (not per-GameObject/fileID).
- Incremental re-index is by mtime; use
--forcefor a fully consistent rebuild.
Development
npm install
npm test # vitest
npm run build # tsc + copy web assets to dist/
This repo also uses an internal agent Harness for development process (see docs/HARNESS.md); it is not part of the shipped package.
License
MIT — see LICENSE.
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.
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.
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.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
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.
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.
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.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases