Security Disclosure Policy

Email a description of the issue, reproduction steps, and any proof-of-concept to [email protected].

Please include:

• A clear summary of what is vulnerable
• Step-by-step reproduction (curl examples are great)
• The impact you believe it has
• Whether the issue is already disclosed elsewhere
• How you'd like to be credited (or to remain anonymous)

We accept reports in English or Spanish. For especially sensitive findings, you may request a PGP-encrypted exchange — we will provide a key on request.

In scope:

• calavai.com and all sub-paths
• /api/mcp/pro — the Model Context Protocol server
• /api/dashboard/* — Pro account surface
• Authentication flows: /api/login, /api/auth/*
• Stripe webhook handler at /api/webhooks/stripe (replay, signature handling)
• Tenant-isolation correctness across every Prisma query
• Sanitization correctness in src/lib/mcp/sanitize.ts
• The Pro MCP API key lifecycle (create, validate, revoke, rotate)

Out of scope:

• Denial-of-service testing against production (you may demonstrate the attack class in writing without flooding our infrastructure)
• Social engineering of Calavai staff or users
• Physical attacks against operator hardware
• Findings that require already-compromised user accounts (Calavai cannot defend against an attacker who controls the user's email and password — that is the user's perimeter)
• Vulnerabilities in third-party services (Stripe, Resend, Google, Vercel, Neon, Upstash, Cloudflare) — please report those to the vendor
• Already-disclosed CVEs in @modelcontextprotocol/sdk when we are on the patched version
• Missing security headers on static pages with no sensitive content
• Volumetric DDoS findings (we rely on Vercel + Cloudflare to absorb these)

If you act in good faith within this policy, Calavai will:

• Not pursue or support civil or criminal action against you under the Computer Fraud and Abuse Act (CFAA), state computer-crime statutes, the DMCA anti-circumvention provisions, or equivalents
• Not threaten the Texas Business and Commerce Code §324 or Texas Penal Code §33 against you
• Not request that internet service providers, hosting providers, or domain registrars take adverse action against you for the research itself
• Consider this written safe harbor an authorization granted to you under all applicable computer-misuse and access-control statutes, to the maximum extent permitted by law

This protection ends if you:

• Exfiltrate or retain user data beyond what is strictly necessary to demonstrate the issue (anonymize quickly, delete promptly)
• Cause harm to users, Calavai, or third parties
• Publicly disclose the vulnerability before we agree on a coordinated disclosure date
• Demand payment to withhold disclosure (extortion)
• Violate the rights of any third party while researching

Our targets, measured from receipt of a well-formed report:

• Initial acknowledgment: within 72 hours (we are a small team and operate primarily Monday–Friday)
• Triage decision (in-scope / out-of-scope, severity assignment): within 7 days
• Critical fix shipped: target 30 days. We will keep you informed if a fix requires longer (e.g., third-party dependency)
• High-severity fix shipped: target 60 days
• Coordinated disclosure: by mutual agreement, default 90 days after fix or initial report (whichever is sooner)

Calavai does notoffer monetary rewards, paid bug bounties, swag, or any other material compensation for reported findings. We are an early-stage product with no revenue to fund a payout pool, and we'd rather promise nothing than promise and not deliver.

What we offer is:

• Public credit on this page (with your consent), linkable from your own profile or LinkedIn — useful for portfolios, conference talks, and employer references

If we ever launch a paid bounty program in the future, prior reporters listed in the Hall of Thanks will be considered first. No timeline is committed.

These reduce friction for both sides:

• One issue per report. Multiple distinct issues = multiple emails.
• If you need a test account, ask. Do not register fake Pros or run automated registration against production.
• If you discover personally-identifiable information of other users, please do not view or copy it beyond what is needed to demonstrate the issue. Anonymize in your report.
• Do not test for issues that require destructive actions (mass cancel, bulk refund, account deletion). Demonstrate in writing instead. We will accept a clear theoretical proof if the supporting evidence is convincing.

Researchers who responsibly disclosed issues to Calavai. Listed with permission.

(We launched our security policy in May 2026. This list is currently empty. Be the first.)