SystemLens

Give ChatGPT trustworthy visibility into the Windows PC it is troubleshooting—and verify what changed afterward.

SystemLens is an open-source, read-only Windows diagnostic companion and MCP architecture by Quod. It collects bounded system evidence, normalizes it into stable schemas, and supports a repeatable diagnose → inspect → verify workflow.

Project status: 0.3.0-showcase is a source-first Showcase/demo release. The deterministic local demo is ready to run; a hosted production plugin service is not currently operated. SystemLens is not created, supported, certified, or endorsed by OpenAI.

Why SystemLens exists

Ordinary troubleshooting is speculative: describe a symptom, try a suggestion, and hope the relevant state changed. SystemLens makes that loop evidence-first:

1. Inspect a bounded set of Windows diagnostic categories.

2. Save a short-lived normalized baseline.

3. Let the user perform the recommended change.

4. Inspect the same categories again.

5. Return deterministic deltas without claiming that correlation proves causation.

The flagship fixture starts with low C: free space, Discord enabled at startup, an older display driver with a problem code, and repeated display resets. The after-state has more free space, Discord disabled at startup, a changed driver, and no new display resets.

Related MCP server: Windows Forensics MCP Server

Try the deterministic demo

The demo is synthetic, sanitized, and clearly labeled. It uses the real fixture loader, snapshot store, and comparison logic; it does not contact a machine, hosted backend, or OpenAI API.

Prerequisites: Node.js 24 and pnpm 11.

git clone https://github.com/Quod-cmd/SystemLens.git
cd SystemLens
pnpm install --frozen-lockfile
pnpm demo

The JSON result contains the before state, the created snapshot, the after state, and structured verification evidence. See the demo guide for the expected proof points.

Architecture

The open-source demo runs locally from deterministic fixtures. The fuller architecture demonstrates how an MCP backend could connect ChatGPT to a Windows companion through a secure outbound device channel. That hosted service is intentionally not operated for this release.

• Companion (companion/) — C#/.NET read-only collectors, local sanitization, explicit capability status, DPAPI credential storage for a future paired deployment.

• MCP server (server/) — TypeScript, Express, MCP TypeScript SDK, strict schemas, OAuth resource-server boundary, device ownership checks, snapshots, and comparison logic.

• Evidence widget (web/) — React/Vite MCP UI for bounded snapshot and comparison evidence.

• Fixtures (tests/fixtures/) — deterministic, invented diagnostic scenarios only.

Read the detailed architecture, tool contracts, and threat model.

Diagnostic categories

SystemLens defines bounded collectors for:

• Windows, CPU, memory, GPU, uptime, and power-plan overview

• fixed-volume capacity, free space, filesystem, and available health metadata

• process names and bounded CPU/memory summaries—never command lines

• startup application state

• driver versions, dates, status, and problem codes

• grouped, sanitized crashes and allowlisted Windows events

• exact or predefined service status

• minimized adapter, gateway, DNS, and internet reachability state

• attached USB metadata without serial numbers

• bounded installed-software searches

• safe OS-reported temperature availability, including explicit unsupported states

Read-only security boundary

SystemLens is an evidence plane, not a remote administration plane.

It cannot run PowerShell or arbitrary commands, read arbitrary files, inspect browser history or credentials, capture keystrokes or screenshots, terminate processes, edit the Registry, reconfigure services or networking, install software, update drivers, or change Windows settings.

Collection is on demand. Results are schema-validated and bounded. User-profile paths and common secret patterns are sanitized locally. Unknown, unsupported, access-denied, and timeout states remain visible instead of being reported as healthy. See the privacy/data map.

Run the full project locally

The deterministic demo and TypeScript tests are cross-platform. The companion and real-machine collector validation require Windows and .NET SDK 10.

pnpm install --frozen-lockfile
pnpm build
pnpm test
pnpm test:companion
pnpm validate
pnpm scan:public

To start the fixture-backed development server:

$env:APP_ENV = 'development'
$env:DEVICE_GATEWAY_MODE = 'fixture'
$env:FIXTURE_SCENARIO = 'flagship-before'
pnpm dev

The local development bearer identifiers are dummy fixtures and are rejected by production mode. No OpenAI API key is required because SystemLens implements MCP and does not invoke an OpenAI model API at runtime.

Real-machine local mode

The companion can run its fixed read-only collector allowlist directly on a Windows PC. That local collection path does not require Render, Auth0, production OAuth, a public MCP server, or an OpenAI account. Remote ChatGPT integration would additionally require a separately operated backend and authentication service; those are architectural components, not a currently hosted product.

Do not distribute the unsigned executable as a mainstream installer. Build from source for local evaluation:

dotnet build SystemLens.sln -c Release
dotnet run --project companion/SystemLens.Companion -- collect system_overview

Validation

The release gate builds the server and widget, runs backend/widget/companion tests, validates contracts, executes the deterministic demo, checks Showcase field limits, scans credentials, and audits the tracked tree plus complete Git history. The exact public-release results are recorded in release notes.

Repository map

companion/       Windows companion, collectors, normalization, transport, tests
server/          MCP/backend, fixture gateway, snapshots, demo, tests
web/             evidence widget and sanitized preview
tests/fixtures/  deterministic invented before/after data
assets/          public cover, architecture, and evidence media
docs/            architecture, demo, security, release, and historical launch audits
scripts/         validation, packaging, secret scanning, public-source audit

Earlier production Plugin Directory work is preserved for engineering history, but that launch track is deferred. The current release is the open-source Showcase edition; it does not promise a hosted backend, consumer installer, or Plugin Directory availability. See the project status and deferred-track note.

Contributing and security

See CONTRIBUTING.md for the intentionally narrow maintenance scope and SECURITY.md for private vulnerability reporting. Never attach real telemetry, credentials, or personal system data to a public issue.

Built with Codex

Codex was used across architecture, Windows collectors, MCP contracts, security hardening, automated tests, the evidence widget, documentation, media preparation, and release auditing. Design decisions and generated work were iteratively checked against executable tests, deterministic fixtures, privacy boundaries, and read-only real-machine validation.

License

SystemLens is available under the permissive MIT License: a simple fit for an inspectable demo and reusable reference implementation. Copyright © 2026 Quod.