Vital Agent Sync handles sensitive health summaries and pairing credentials. Please do not publish vulnerabilities, credentials, onboarding links, QR codes, databases, logs containing private data, or proof-of-concept payloads in a public issue.

Security fixes are provided for the latest released vitalmcp version and the current main branch. Older previews may not receive patches.

Use GitHub's private vulnerability reporting flow for this repository. Include:

• the affected version or commit;
• the affected transport (lan, tailscale, hosted relay, or self-hosted relay);
• reproduction steps using synthetic data;
• the expected and observed security boundary;
• a suggested fix, if available.

Do not include real Apple Health exports, live pairing artifacts, private keys, access tokens, or a user's SQLite database. If a report cannot be reproduced without sensitive material, first describe the minimum metadata required and wait for a private response.

• LAN and Tailscale terminate at the user-owned Vital Agent Sync receiver.
• Health summaries are stored in the user's local SQLite database by default.
• Pairing and relay credentials remain local and must not be pasted into Agent chats, logs, issues, or support messages.
• An Agent or model provider may receive the scoped health context returned through MCP. Local storage does not imply local model inference.
• Vital Agent Sync is not a medical device and must not be used for diagnosis, treatment, or emergency decisions.

apps/www and apps/pitch are exported as static artifacts. Next.js, React, PostCSS, Sharp, and the other site framework packages are development-only build tools; they are not installed or executed in the deployed static-site runtime. Builds consume repository-controlled source and media only and must not process untrusted stylesheets, source maps, images, or other inputs.

The root development dependency tree may temporarily report upstream Next.js build-tool advisories even when audit:www-dependencies and audit:pitch-dependencies report zero production vulnerabilities. This is a narrowly documented exception, not a blanket audit waiver. Maintainers must review those advisories for each release and update to a stable patched framework release when one is available.

See the direct LAN security model, the relay threat model, and the privacy boundary for details.