MCP Documentation · Content Policy
Content Policy
Template Vault generates real, deployable websites. To keep the platform safe and useful for small-business owners, the generate_site tool declines requests in the categories below. Both the MCP server's pre-filter and the underlying Anthropic Sonnet model enforce this policy.
Prohibited categories
Requests obviously matching any of these categories are rejected synchronously with an invalid_request error pointing to this page. Each pattern is a multi-word phrase (not a single ambiguous word), so legitimate businesses with category-adjacent names are not blocked.
Adult / sexual content
- Pornography sites, escort services, sexual webcam services, fetish marketplaces.
- OnlyFans-management agencies acting as redirectors.
- Not blocked: mainstream adjacent businesses — dating apps, swimwear retailers, fertility clinics, sex education for teens, pole-fitness studios, adult learning centers.
Hate speech & extremism
- White-supremacist / nationalist sites, neo-Nazi material, KKK-affiliated content, Holocaust denial, racial-purity advocacy, ethnic-cleansing advocacy.
- Not blocked: Holocaust memorials, anti-racism education, historians or journalists writing about extremism.
Scams & financial fraud
- Ponzi / pyramid schemes, get-rich-quick sites, advance-fee fraud, fake-invoice / fake-bank phishing landing pages.
- Not blocked: legitimate MLM companies, recovery-of-funds law firms, financial-fraud education.
Illegal goods & services
- Sites selling controlled substances without authorization, unlicensed firearms, stolen goods, hacking-for-hire.
- Not blocked: dispensaries operating in legal jurisdictions, FFL-licensed firearms dealers, legitimate cybersecurity consultancies, addiction-recovery clinics.
Child safety
- Any content sexualizing minors (CSAM), grooming materials, content designed to harm children.
- Not blocked: child-services nonprofits, pediatric clinics, kids' activity centers.
Impersonation / phishing
- Sites pretending to be a bank, government agency, or specific brand without authorization.
- Credential-harvesting phishing kits.
- Not blocked: fan sites, parody / satire sites that are clearly labeled.
How rejections work
When the pre-filter matches a prohibited category, the tool returns:
Template Vault declines to generate websites for businesses in this category. The request matched: "<phrase>" (<category>). If you believe this is a false positive (educational, memorial, recovery, news, or research context), please email contact@the-template-vault.com — we review manually within 1 business day. Full policy: https://www.the-template-vault.com/docs/mcp/content-policy
The calling agent (Claude, ChatGPT, Cursor) should surface this message verbatim to the user. The category and matched phrase are logged for audit purposes; the raw business description is not stored.
Defense in depth
Even when the pre-filter misses something (rare wording, euphemisms, intent expressed in structure rather than words), the underlying Anthropic Sonnet model applies its own content safety layer. A request that passes the phrase filter but is still harmful will typically be refused by the model before any site is created. The pre-filter is for speed + clarity; the model is the deeper safety net.
Appeals — false positives
The phrase pre-filter can occasionally false-positive — for example, a Holocaust memorial nonprofit, an addiction-recovery clinic, a sex-education program, an anti-fraud watchdog. If you believe your legitimate business was incorrectly declined:
- Email contact@the-template-vault.com with the business name and a short description.
- We review manually within 1 business day.
- Confirmed false positives result in a refined regex (your business + similar cases get through next time).
Reporting violations
If you've encountered a Template-Vault-generated site that you believe violates this policy, email contact@the-template-vault.com with the URL. We act on reports within 1 business day, which usually means taking the site offline and revoking the owner's generation quota pending review.
Source
The exact phrase patterns the pre-filter uses are open source and reviewable at github.com/julioMeif/template-vault-mcp/CONTENT_POLICY.md. Pull requests refining the patterns (especially documented false-positive cases) are welcome.