12 Free Unity MCP 1.5 Checks Before You Wire Agents - 2026

If you searched unity mcp checks after Unity-MCP 1.5.0 landed on 24 October 2026, you do not need another “agent builds your whole game” trailer. You need a numbered free preflight kit before Cursor, Claude Code, Copilot, or Gemini gets write tools against a Unity Editor that still owns your shipping scenes.

IvanMurzak’s Unity-MCP is the open bridge (skills, MCP tools, CLI). 1.5.0 specifically pushed standalone CLI/App skills, platform-aware executables, headless EditMode CI, and a firmer tests-run lock — mapped in What Unity MCP 1.5.0 Changed for AI Game Dev. Shared bits also show on NuGet (Common 1.5.0). Missing from the SERP is a Discord-pasteable pass that forces lab quarantine, OS-correct skills, read-only smoke, and NO_LIVE_AGENT before anyone wires main.

This listicle owns the free kit. It is not the 1.5.0 what-changed map. It is not Unity MCP + Cursor first safe bridge (first-night ritual). It is not Meshy Blender MCP checks (DCC props). It is not Unreal MCP loopback Stop. This URL owns Unity-MCP 1.5 enablement checks before agent write.

Why this matters now

1.5 made skills and CI less fragile. That increases FOMO: “upgrade, connect Cursor, let it refactor overnight.” The blast radius did not shrink — an agent that can invoke Editor tools can still trash scenes, fight test runners, or emit wrong-OS shell commands on a mixed Windows/macOS team. Free checks are the cheapest insurance.

Direct answer: Run checks 1–4 before you install 1.5 on anything that matters. Run 5–9 on a disposable lab project with read-only tools first. Run 10–12 before write tools may touch a branch that ships. If check 11 fails (no NO_LIVE_AGENT latch), stop — that is the point of the kit. Authority: 1.5.0 release + current repo README.

Who this list is for

Audience	Use it to…
Beginners	Survive first MCP wire without nuking `main`
Developers	Skills OS, test lock, headless vs render honesty
Creators	Paste a twelve-check card in Discord
Companies	Diligence — third-party MCP pin + write-tool policy

Time: 25 minutes for 1–4; 2–3 hours for a full lab smoke.
Honest limit: Unity-MCP is third-party open source — not a Unity Technologies product stamp unless their docs say so. Re-read the tag when you bump.

Non-repetition

URL	Job
Unity MCP 1.5.0 what-changed	Release aftermath map
Unity MCP + Cursor first bridge	First-session ritual
Unity CLI first agent-safe session	CLI agent evening
Meshy Blender MCP 12 checks	Different MCP stack
Unreal MCP loopback Stop	Unreal network refusal
Godot AI MCP first session	Godot MCP evening
Unity guide	Engine hub
25 free Discord servers	Paste channel
This URL	Numbered free Unity MCP 1.5 checks

Four-check minimum (if you only have twenty-five minutes)

#	Check	Fail means
1	Read 1.5.0 release + what-changed once	Myths about “official Unity AI”
2	Disposable lab project path ready	Shipping `main` Editor
3	Plugin + server + Common versions aligned	Half-upgrade soup
4	NO_LIVE_AGENT written for shipping projects	Overnight write into release

Do not enable write tools on a shipping Editor until these four are green.

The twelve checks

1. Docs latch (1.5.0 + map)

Do: Open the 1.5.0 release and skim What Unity MCP 1.5.0 Changed. Note skills (#1063), platform commands (#1064), headless EditMode (#1061), tests-run lock (#1049).

Pass: Ticket links both.
Fail: “Stars are enough.”
Next: Keep the release URL in the agent prompt footer.

2. Lab project quarantine

Do: Clone or copy a throwaway Unity project (UnityMcp15-Lab-YYYYMMDD). Do not point the MCP client at the only shipping repo. Separate Library/cache if your install path is global — document which Editor instance is lab.

Pass: Lab path on the receipt; shipping project not in the client’s default workspace.
Fail: “We’ll revert git later.”
Next: Same instinct as first Cursor bridge — copy first.

3. Topology pin (Client / Server / Plugin / Common)

Do: Record versions for MCP client, Unity-MCP server, Editor plugin, and any Common NuGet/package. Bumping only Common while leaving the plugin on 1.4 is a fail.

Pass: Four version strings on the receipt.
Fail: “Whatever Package Manager resolved.”
Next: Re-read NuGet/repo topology notes when confused.

4. NO_LIVE_AGENT latch

Do: Write an explicit policy: shipping branches / release Editors have write tools disabled (or MCP disconnected) until checks 5–12 pass and a human signs promote. Soft sticky notes count; verbal “be careful” does not.

Pass: NO_LIVE_AGENT=yes on shipping; lab may enable writes later.
Fail: Same Cursor window can write either repo.
Next: Company diligence four-liner from the what-changed post.

5. Install 1.5 on lab only

Do: Follow current repo README install (CLI quick setup / packages — not a 2025 mirror). Confirm Editor opens with the bridge healthy on the lab project.

Pass: Lab Editor shows expected MCP/plugin state.
Fail: Installed onto shipping project “just to save time.”
Next: Screenshot About/package versions into the receipt.

6. CLI / App skills visible

Do: Exercise 1.5’s standalone CLI/App skills path enough to list or generate skills with global metadata. You are proving the headline feature exists in your install.

Pass: Skills list or generate step logged.
Fail: “We’ll use chat tools only; skills later.”
Next: If skills missing, you are not on 1.5 — stop and re-pin.

7. Platform-aware command match

Do: On each OS your team uses (at least your machine + one peer OS if mixed), confirm skill-generated executables/shell lines match the host (PowerShell vs bash, .exe vs no extension, path separators).

Pass: Commands look runnable without translation.
Fail: macOS skill blobs pasted into Windows CI.
Next: This is the #1064 reason 1.5 matters — treat as a gate, not a nice-to-have.

8. Read-only smoke before write tools

Do: First agent session: read-only tools only (inspect hierarchy, search, explain). No create/delete/refactor until a human OK. Log the tool allowlist.

Pass: Session transcript shows read-only.
Fail: “Make a menu” as first prompt.
Next: Pair with first-bridge article habits for juniors.

9. Tests-run lock awareness

Do: Know that 1.5 holds the tests-run lock until Unity’s run finishes and frees on error. In lab, avoid launching overlapping EditMode runs from agent + CI simultaneously. Document whether agents may call “run tests” at all.

Pass: Policy: serialize test tools / one runner.
Fail: Agent spam-clicks Test Runner while CI is hot.
Next: Flakes after upgrade often mean lock policy ignored, not “Unity is broken.”

10. Headless EditMode ≠ render done

Do: If you adopt EditMode -nographics CI from the 1.5 wave, label the job honestly. Require a separate rendering / Play Mode smoke before art or camera changes ship.

Pass: Pipeline diagram shows two legs.
Fail: “Headless green — ship the build.”
Next: Quote this line in Discord when producers collapse the legs.

11. Write-scope + blast radius

Do: Cap agent session: max N mutating tool calls, no scene deletes outside Lab/ folders, no Package Manager upgrades without human, stop on first unexpected compile storm. Prefer feature-branch lab clones over shared develop.

Pass: Written agent policy with N (e.g. 5) and stop rules.
Fail: Overnight “fix everything.”
Next: Same spirit as Unreal MCP discipline — different stack, same blast radius.

12. Promote gate (lab → shipping pin)

Do: Tech lead signs: checks 1–11 green, versions pinned in source control, OS matrix noted, write tools still off on shipping until a second human ack, changelog entry for the MCP bump. Only then update the shipping project’s Unity-MCP pin — still prefer read-only for the first shipping day.

Pass: Signed promote receipt.
Fail: “It worked in lab — bump everyone.”
Next: Schedule a calm week, not a content milestone that depends on unrestricted agents.

Full lab receipt (copy/paste)

UNITY_MCP_1_5_CHECKS_RECEIPT_v1
date:
operator:
lab_project_path:
shipping_project_path: (must stay unwired)

check_01_docs_latch: PASS/FAIL
check_02_lab_quarantine: PASS/FAIL
check_03_topology_versions:
  client:
  server:
  plugin:
  common:
check_04_no_live_agent: PASS/FAIL
check_05_lab_install_1_5: PASS/FAIL
check_06_cli_app_skills: PASS/FAIL
check_07_platform_commands:
  os_tested:
  peer_os_tested: yes/no/na
  match: PASS/FAIL
check_08_read_only_smoke: PASS/FAIL
  allowlist_note:
check_09_tests_run_lock_policy: PASS/FAIL
check_10_headless_vs_render_split: PASS/FAIL
check_11_write_scope:
  max_mutating_calls_N:
  stop_rules:
check_12_promote_signed_by:
  date:

blockers:
next_review_date:

Fill every line. Blank topology versions = automatic fail on check 3.

How the twelve map to 1.5 themes

1.5 theme	Checks that prove you absorbed it
CLI/App skills + metadata	6
Platform-aware executables	7
Tests-run lock	9
Headless EditMode / render split	10
“Upgrade FOMO” generally	1, 2, 4, 5, 12
Everyday agent blast radius	8, 11

If you only green checks that feel like “install worked” (5) and skip 7/9/11, you upgraded the package without upgrading your ops — the failure mode 1.5’s release notes cannot fix for you.

Excuse ladder (kill these in standup)

Excuse	Why it fails	Point at
“We’ll only ask nice prompts.”	Prompts are not allowlists	8, 11
“CI will catch agent mistakes.”	CI is not a scene undo stack	2, 4
“1.5 fixed locks so spam tests.”	Lock ≠ license to parallelize forever	9
“Headless is enough for art.”	Wrong leg	10
“Meshy checklist is close enough.”	Different stack	Non-repetition
“Stars mean secure.”	Popularity ≠ pin policy	1, 3, 12
“Junior can wire shipping; I reviewed the PR title.”	Review the receipt	12

Check card template

Check #:
Owner:
PASS / FAIL / N/A:
Evidence:
Blocker if FAIL:

N/A without a reason is a fail. Screenshot evidence beats “I remember it worked” every time on review day — keep the files.

Beginner evening path

Checks 1–4 on paper (25–40 min).
Lab install 1.5 (check 5) (30–60 min).
Skills + OS command verify (6–7) (20–40 min).
Read-only smoke (8) (15–30 min).
Write policy notes for 9–11; leave 12 for lead tomorrow.

If anything in 5–8 fails, do not invent workarounds on the shipping repo — fix lab first.

Beginner “done” definition

You are done for the night when: the lab project opens with 1.5.0 pieces aligned, you have a skills list or generate proof, one read-only agent turn exists in the transcript, and the shipping project’s MCP client still cannot mutate it. You are not done when the Editor merely launches without errors — that is only check 5 adjacent.

Developer depth — optional lab proofs

Proof	How	Pass
Lock serialization	Start EditMode; attempt second run	Second waits/fails cleanly — no corrupt overlap
Wrong-OS negative	Paste peer OS command on purpose	Fails predictably; document
Allowlist	Client config export	Write tools absent until check 12
Version grep	`manifest` / packages lock in git	1.5.0 strings committed on lab branch

KEEP lab / HOLD ship (body)

Decision	When
KEEP lab	Any fail in 1–11; first write tools; CI redesign
HOLD ship	1–12 green; pin reviewed; NO_LIVE_AGENT still true on release Editors until deliberate enable

Failure modes

Failure	Symptom	Fix
Half topology	Common 1.5, plugin 1.4	Align all pieces (check 3)
Wrong-OS skills	Commands die on peer machines	Check 7 on each OS
Write day one	Scene thrash	Check 8 then 11
Parallel tests	Flaky EditMode	Check 9
Headless folklore	Art bugs ship	Check 10
Meshy kit copy-paste	Triangle budgets on Unity ticket	Wrong stack — use this kit
Stars diligence	“4550 stars = safe”	Checks 1 + 4
Promote without pin	Snowflakes per machine	Check 12 + git lockfiles
Shared develop lab	Two agents, one Library	Per-dev lab clones
README drift	Old install steps	Current repo README only

Company / diligence four-liner

Stack: Third-party Unity-MCP 1.5.0 (Apache-2.0) — release.
Control: Twelve free checks + NO_LIVE_AGENT before write tools on shipping Editors.
CI honesty: Headless EditMode ≠ render certificate.
Evidence: Version pins in source control + signed promote receipt (check 12).

Discord-ready brief (paste)

Unity-MCP 1.5 checks before wiring agents:
1–4 docs + lab + topology + NO_LIVE_AGENT
5–8 install lab, skills, OS commands, read-only smoke
9–10 test-lock policy + headless ≠ render
11–12 write-scope + signed promote
Don’t bump shipping until the card is green.

What to do Monday

Block	Minutes	Outcome
A	25	Checks 1–4 on paper
B	45	Lab 1.5 install (5)
C	30	Skills + OS (6–7)
D	20	Read-only smoke (8)
E	20	Policies 9–11 drafted
F	10	Lead schedules check 12

If Monday is interrupted after block C, still paste the Discord brief and leave shipping Editors disconnected. Partial installs on main are worse than a delayed lab night — resume Tuesday at check 5 with a clean receipt rather than “finishing” on the wrong project.

Key takeaways

Twelve free checks gate Unity-MCP 1.5 agent wiring — not vibes.
Lab quarantine before shipping Editors.
Align client / server / plugin / Common versions.
NO_LIVE_AGENT is a latch, not a mood.
CLI/App skills must exist on your 1.5 install.
Platform-aware commands must match each host OS.
Read-only smoke before any mutate tool.
Serialize tests — respect the 1.5 run lock.
Headless green ≠ render done.
Write-scope caps stop overnight factories.
Promote is a human signature, not a Slack shrug.
This kit ≠ Meshy / Unreal MCP kits — different blast radii.
Fill UNITY_MCP_1_5_CHECKS_RECEIPT_v1 — blank topology lines fail check 3.
Map vs kit: what-changed explains 1.5; this URL is the pass/fail card.

FAQ

What are Unity MCP checks?

A free preflight list before enabling AI agents (via MCP clients) against Unity through IvanMurzak’s Unity-MCP — especially after the 1.5.0 skills/CI wave.

Do I need 1.5.0 for this kit?

The kit is tuned for 1.5 headlines (skills, platform commands, headless EditMode, tests-run lock). On older pins, still run lab quarantine + NO_LIVE_AGENT; re-check release notes when you bump.

Is Unity-MCP official Unity software?

Treat it as third-party open source unless Unity’s own documentation says otherwise.

Can I skip read-only smoke if I trust the model?

No. Check 8 exists because trust does not undo a deleted prefab.

Why test skills on every OS?

1.5 generates host-platform executables. Mixed Windows/macOS teams fail silently when one OS’s skill blob is assumed universal.

How is this different from the what-changed article?

What-changed explains why 1.5 matters. This listicle is the pass/fail enablement kit.

How is this different from Meshy Blender MCP checks?

Meshy/Blender is a prop generation DCC loop. Unity-MCP drives Editor/agent loops. Do not copy triangle budgets into Unity tickets as a substitute for NO_LIVE_AGENT.

What should companies require?

Pinned versions in git, NO_LIVE_AGENT on release Editors, signed check-12 promote, and an honest split between headless EditMode and render/Play Mode CI.

Where do I learn the first Cursor session?

Unity MCP with Cursor - first safe editor bridge session.

What if check 7 fails on a teammate’s Mac?

Do not “fix” by hand-editing commands into tribal wiki forever — regenerate on that host and file a bug upstream if 1.5 still emits wrong-OS lines.

Can agents run tests freely after 1.5’s lock fix?

Only under a serialized policy (check 9). The lock reduces corruption; it does not make parallel spam wise.

What do I paste in Discord?

Use the Discord-ready brief above.

Do I need both this kit and the first Cursor bridge article?

Yes if the human has never bridged MCP: bridge article for session mechanics, this kit for 1.5 enablement gates. Juniors often skip quarantine; leads often skip OS command checks — assign both.

What belongs in git after a lab pass?

At minimum: package/manifest pins proving 1.5.0, a short UNITY_MCP_1_5_CHECKS_RECEIPT_v1 (or ticket link), and CI job names that show headless vs render legs if you claim check 10.

Can Marketplace or Asset Store MCP clones substitute?

Only if you re-run the twelve against that fork’s versions. This kit is written against IvanMurzak Unity-MCP 1.5.0 semantics — name swaps without re-verification are a diligence fail.

When do we turn write tools on for shipping?

After check 12, on a feature branch, with N caps from check 11, and preferably a second human on call. Day-one shipping pin can stay read-only even after promote.