Ask. Measure. Ship.
A local MCP server for Unity — it gives your AI client (Claude, Codex, Gemini) evidence-first build, memory & performance audits. Or run it yourself in the Editor. Read-only by default; nothing changes without your confirmation.
Your build is 120 MB. Do you know why? Release Vitals measures what’s actually shipping — and labels every finding by how proven it is, so you act on facts, not guesses.
One prompt each.
Five release workflows instead of hundreds of raw editor commands — the agent gets answers, not chores.
Build Size Audit
Builds the active target and measures it via Unity’s BuildReport — total size, biggest assets, package risks verified against real scene usage, baseline compare on every re-run. Android adds deep APK analysis (apkanalyzer / dexdump / aapt2).
Memory Audit
Guided captures from the Editor or a connected player (on-device via adb on Android), with explicit pause-and-confirm. Asset-level findings — audio decompression, oversized textures, Read/Write flags — tiered by build inclusion.
Performance Audit
Lightweight Editor Play-Mode capture: worst frame windows, GC pressure, active-scene script candidates. Unproven causes are labeled “unattributed,” never invented.
Release Diagnostics & Preflight
Readiness blockers, settings risks, tool detection, logs and evidence bundles — a clean go/no-go before you cut a release.
Build Failure Investigation
A guided log-and-evidence flow for when the build won’t even finish — collect, correlate, and pinpoint the blocker.
Works without AI
Every workflow is available from the Release Doctor window inside Unity, with full run history and artifacts. The AI client is the fast path, not a requirement.
Evidence, not guesses.
Evidence tiers
Every finding labeled artifact-proven, build-confirmed, likely, or unconfirmed. Missing evidence is shown as missing.
Workflows, not a tool bag
Five release goals, not hundreds of raw commands. The agent gets answers, not chores.
Local MCP, by design
One readable JS stdio relay, localhost-only bridge — no daemons, no npm install, no cloud.
Token-economical
Summary-first payloads — a full audit fits inside a free-tier model’s daily budget.
Asks before it acts
Play-Mode gates, build-confirmation flags, explicit capture confirmation. You stay in control.
Your AI client — or you.
Release Vitals is a local MCP server — Model Context Protocol, the open standard that lets AI agents call real tools. It runs on your machine over stdio (no cloud); add it to your client once and your agent runs the audits.
Add it once, then ask
Add the MCP server to Claude, Codex, or Gemini with one command. Then ask in plain language and iterate: ask → measure → fix → re-measure.
✓ release-vitals connected · 15 tools
› audit the android build size
✓ build-confirmed 126.8 MB · top: textures 41%
# re-run to compare against baseline
Release Doctor, in the Editor
Open Tools › Release Vitals › Release Doctor. Every workflow runs from the window with run history and artifacts — no AI required. Same engine, same evidence model underneath.
[ Build Size ] [ Memory ] [ Performance ]
● Bridge running · localhost · read-only
history: 12 runs · artifacts saved
What you need.
Platform support
Every workflow runs from the Editor and is tested on Android, WebGL, Windows & macOS. Build size measures any target via BuildReport; Android adds deep APK analysis and on-device memory capture.
Requirements
- Unity Editor open while AI workflows run
- Node.js 18+ for MCP/AI features — Editor diagnostics work without it
- An MCP-capable client: Claude Code, Codex CLI, Gemini CLI, Cursor — or any MCP stdio client
- Android Build Support for APK analysis & on-device memory
Measure it. Then ship it.
A local MCP server for evidence-first Unity audits — local-first, read-only by default, driven from the AI client you already use.
Extension Asset, per seat · on the Unity Asset Store