Documentation

Scan and optimize your Unity project from the conversation you're already working in.

Scan and optimize from chat

Once connected, your coding agent can scan → plan → apply → re-scan without opening PerfLint's panels. Keep Unity open; ask for the result you want in chat.

“Use PerfLint to scan this project and apply the supported low-risk batch fixes. Leave trade-offs and asset merges for my review. Re-scan and show what changed and what remains.”

Commit your project first. Batch execution requires Pro. The agent can inspect affected code and check preconditions before applying fixes; its own permission settings still apply.

• Automatic settings fixes: PerfLint applies supported import/project settings changes and reports fresh scan results.
• Agent-written code: on either tier, the agent can submit C# or shader edits for compilation verification and automatic rollback on failure. This uses your agent's model, not PerfLint AI credits.

Re-scanning verifies findings, not FPS gains. Runtime sampling and before/after measurements still need the editor workflow; the agent can read saved runtime findings.

Connect your agent

Install PerfLint, then connect using Unity 6, Unity CLI, com.unity.pipeline and an open editor. Configure the CLI's MCP server for your client:

unity mcp configure --list
unity mcp configure claude-code --project-path /path/to/YourProject

Replace claude-code with your client from the list and restart its session. An agent with terminal access can also run PerfLint's CLI commands directly.

["mcp", "--project-path", "/path/to/YourProject"]

[mcp_servers.unity]
command = "/absolute/path/to/unity"
args = ["mcp", "--project-path", "/path/to/YourProject"]
startup_timeout_sec = 120

Requirements

Unity 2021.3 or newer, including Unity 6. Windows, macOS and Linux editors. Scanning is local, free and requires no account.

Install

Open Window ▸ Package Manager ▸ + ▸ Add package from git URL and paste:

https://github.com/cnxiaom/perflint-unity.git

Append #v1.7.2 to pin that release.

Asset Store: get PerfLint, then import it from Package Manager ▸ My Assets. It installs under Assets/PerfLint.

"com.perflint.unity": "https://github.com/cnxiaom/perflint-unity.git"

Scan

1. Open Tools ▸ PerfLint ▸ Scan Project (Ctrl/Cmd + Alt + L).
2. Click Scan Project. Filter by severity or search for a rule, asset or script.
3. Read the finding's evidence and recommendation. Use Locate where available to jump to the source.

If prompted, enable script analysis, wait for compilation, then scan again. Use Export CSV or Export Report to share results. See the rule reference for individual checks.

Runtime Profiler

1. Open Tools ▸ PerfLint ▸ Runtime Profiler (Ctrl/Cmd + Alt + M).
2. Enter Play Mode, reach the scene you want to measure and let loading settle.
3. Click Start Sampling, reproduce the workload, then Stop & Analyze.
4. Wait for hotspot processing, then inspect the findings and use Locate where available.

Deep Profile adds script detail and overhead. Follow any recompile prompts and sample again. Turn it off for representative timing; compare runs with the same profiling settings.

Autopilot

Open Tools ▸ PerfLint ▸ Autopilot, then follow the four steps:

1. Where you are: scan, set target FPS, and choose the Boot from and Measure scenes.
2. Baseline: measure before editing. Calibration runs automatically; navigate to the target scene if needed and let it load.
3. This round: review evidence and risk, select fixes, and apply. Wait for imports and compilation to finish.
4. Did it work: repeat the same scene, camera, workload and settings against the saved baseline.

No measurable change means the difference is within measurement noise. An editor improvement still needs a target-device check.

Apply fixes

Commit your project first and review each fix's preview and scope.

• One-click and batch fixes (Pro): apply supported settings changes. Edit ▸ Undo does not restore these; use version control.
• AI Fix (Free & Pro): generates a focused code edit, verifies compilation and rolls back failed edits.
• Migration Assistant (Pro): previews a whole-file diff and verifies compilation after applying it.

A clean compile does not prove gameplay is correct. Test the affected behavior after any code change.

AI setup

Hosted: Explain and AI Fix work without an API key. Free includes 10 credits/day; Pro includes 5,000/month.

Your own key (either tier): open LLM ▸ Advanced… and add a Claude or DeepSeek key. Requests go directly to that provider at its rates and do not use PerfLint credits.

CLI & CI

With the editor open and Unity CLI set up:

unity command perflint_scan
unity command perflint_list_findings --domain runtime
unity command perflint_fix --dry_run

The runtime query reads saved measurements; it does not start sampling. Preview fixes before applying them. Applying batch fixes requires Pro.

Unity -batchmode -nographics -projectPath . \
  -executeMethod PerfLint.Ci.PerfLintCli.RunGate \
  -perflintMaxCritical 0 \
  -perflintReportHtml perflint-report.html -logFile -

name: PerfLint health gate
on: [pull_request]
jobs:
  gate:
    runs-on: [self-hosted, unity]
    steps:
      - uses: actions/checkout@v4
      - shell: bash
        env:
          UNITY: /path/to/Unity
        run: |
          "$UNITY" -batchmode -nographics -projectPath . \
            -executeMethod PerfLint.Ci.PerfLintCli.RunGate \
            -perflintMaxCritical 0 \
            -perflintReportHtml perflint-report.html -logFile -
      - uses: actions/upload-artifact@v4
        if: always()
        with:
          name: perflint-report
          path: perflint-report.html

Activate Pro

Click Free / Pro in the PerfLint toolbar, paste your license key and click Activate. Deactivate on a machine to free its seat.

See plans and pricing for a license.

Privacy

• Scanning and sampling: local; no project upload.
• Explain and AI Fix: selected finding metadata or code goes to the configured AI service. Hosted requests use PerfLint's proxy; your own key connects directly.
• Connected agents: requested findings include asset paths and line numbers and follow your agent's data policy.
• Licensing: activation and periodic validation contact the license service.

See the privacy policy.