Carbon Lite MCP Server

CARBON Lite — local skills and optional background checks

The eight CARBON Lite workflows plus a portable Studio workspace. Open source under MIT, readable JavaScript. The Codex plugin needs no MCP server, OAuth, account, analytics, hosted service or CARBON model calls. Built by testers.ai. This is a separate edition; it does not replace the existing Claude package or the original Studio submission. The separate private CARBON Studio Pro edition preserves the full local MCP experience. Lite is the first package intended for submission.

Standalone MCP distribution

For MCP clients and directories, use the separate local MCP bundle. It adds stdio tools and prompts around the Lite evidence engine, not Pro features. The Codex plugin ZIP remains serverless.

Install in Codex

Download the ZIP from Releases, extract it, and use your Codex version's local-plugin installation flow to select that folder. This repository is not a claim of public directory approval. After installation, start a fresh conversation and run /carbon (select CARBON Lite's carbon command in the host's skill picker if it requires a plugin prefix). CARBON tests the current project when the target is clear. Otherwise it asks what to test: provide a URL, folder, repository, API, app, feature, or requirements. It runs a broad assessment within the configured budget and reports findings, evidence-qualified confidence, coverage gaps, and prioritized next steps. Host capabilities and permissions determine which tests can actually run.

License

CARBON Lite is open source under the MIT license. You may use it at work, fork it, modify it, redistribute it, and sell copies, preserving the required copyright and license notices. Third-party dependency licenses remain intact; see NOTICE.md. There is no trademark endorsement grant.

Commands

Invocation prefixes depend on the host. A useful first request is: “Use CARBON Lite to assess this project and show the evidence report.”

Skills can analyze supplied requirements or results without a local runtime. Hands-on testing needs the host's browser/terminal/API capabilities. The optional evidence helper requires Node 22+, already bundled as readable JavaScript: no npm install for archive users. Demo artifacts may require Python or their documented test framework; creating a copy neither installs dependencies nor starts a server.

What is and is not equivalent

Included: the Light investigation methods, four disposable demos/two test profiles, local preferences, revisioned evidence/history, findings, persona journeys, screenshots, scope-qualified confidence, portable HTML/JSON, screenshot enlargement, map/check inspection and explicit feedback import/export.

Different from MCP Studio: no native global/settings entrypoint, MCP tools, automatic host message handoff, unattended UI write-back, or guaranteed in-chat live rendering. The report is a normal HTML artifact. Supporting browsers can watch a user-selected snapshot file; others need reload/import. UI feedback is a draft until imported. Not included: Pro commands, proprietary benchmarks, hosted sharing, cloud execution, continuous loops or cross-build/3D map features. This is not full native-UI parity.

Cost and privacy

No infrastructure bill for this edition. Host subscriptions/API use, local compute and any test-target costs remain separate. State stays in the selected project's .carbon/studio-lite/; no hidden account or global configuration is created. Outputs are private local files but contain any evidence recorded. Review before sharing. The host model provider's processing is separate from this plugin. No automatic retention/deletion: remove only the specific project evidence folder when you choose. Never remove other CARBON state as part of uninstalling this edition.

Develop and package

npm ci
npm run package

The ZIP contains the manifest, skills, runnable helper, full source, bundled testing reference catalogs, fixture code and third-party notices. It excludes node_modules, local evidence, test outputs, credentials and private benchmark data.

See SUBMISSION.md for the local skills/hooks review path and remaining portal checks. Building a ZIP is not marketplace acceptance or publication.

Talk to Jay

Use /jay or /j to talk directly to Jay · AI Test Manager Agent. The MCP prompt router also accepts J. Skill names are lowercase; if a host does not resolve /J, use /j. Claude commonly prefixes these as /carbon:jay and /carbon:j; Codex may show a plugin-prefixed command or $jay in its skill picker. The plugin cannot reserve a global slash name in every host.

• /jay What should we test next? — discuss risks and the evidence we have.
• /j Test this project — run this edition's normal CARBON assessment.
• /jay Explain the last report — interpret recorded results without rerunning.
• /carbon remains the direct broad-testing command.

Jay replies in first person, in a frank, practical tone inspired by Jason Arbon. He is an AI persona, not Jason. With no clear testing target he asks for a URL, folder/repository, API, app/feature, or requirements. He does not create a demo unless asked. Questions do not automatically start tests, and fixes still need authorization. Lite keeps its existing capabilities; these are two conversational entrypoints, not two new testing workflows or an upgrade to Pro.

Jay's bundled icon appears in compatible command pickers and CARBON report views. Host chat avatars and image sizing are host-controlled; text-only chat uses “Jay · AI test manager” instead of a large image.

Optional Jay background checks

After an explicit per-project choice and host hook trust, Jay checks changed code during breaks. Defaults: 2-minute idle delay, 3-minute runs, maximum 10 minutes per run, one worker, 3 runs and 10 reserved minutes per UTC day across projects. No automatic fixes or live-site testing. Local checks and outcome counts stay on this machine. Optional Claude-only AI source review requires separate consent to provider processing and allowance use. Codex checks do not launch Claude. Use carbon-background to enable, pause, review, or disable everywhere. See behavior, controls and privacy. Unsupported hosts retain manual testing.