Claude

Connectors Directory. Team or Enterprise org. Portal is in Claude.ai settings.

Shared

Our product call, not a store rule: bash_exec stays on tools/list.

Note
Reviewer demo login if the server requires sign-in. No MFA, SMS, or email-confirm. Claude also wants a fully populated account and every URL/step written out.

Human

Create the reviewer login. ChatGPT only requires this if OAuth/sign-in is on. Claude requires test credentials.
HTTPS website, support, privacy, and terms URLs. Live: sails.app, /contact, /legal/privacy, /legal/terms.

AI
Our copy lock, not a store rule: 0 customer sentences say Sails makes zero inference calls.

Note

Tool responses omit secrets, debug lines, internal ids, and user fields the privacy policy does not disclose. Privacy policy matches what tools return.

Reported 2026-08-28.

Official ChatGPT rule: strip unnecessary personal data, auth secrets, debug payloads, internal identifiers, and undisclosed user-related fields from tool responses. Then make https://sails.app/legal/privacy match what remains. Not a field-by-field inventory for its own sake.

ChatGPT: every tool has readOnlyHint, openWorldHint, destructiveHint, and a justification for each. Claude: every tool has a title and readOnlyHint or destructiveHint.

Reported 2026-08-28.

Platform work on the MCP server behind https://sails.app/mcp. ChatGPT requires all three hints plus a justification string for each value. Claude requires a title and readOnlyHint or destructiveHint. Bits must match side effects. Do not hide bash_exec. If you do not have the MCP server source, stop.

Claude

v1 listing is community. Verified is a later Anthropic pick. 0 extra product work.

Note

Authenticated services use OAuth 2.0. Portal also allows custom connection or no auth if that is how Sails works.

Confirm what https://sails.app/mcp actually does on initialize (401 + WWW-Authenticate vs open). Document the real mode. Do not invent PKCE work if the portal accepts the current mode. If you do not have the MCP server source, stop.

7 portal acknowledgments: guidelines, first-party API, financial, AI media, prompt injection, conversation data, public docs.

Human

You check these in the Claude portal.
Privacy URL, documentation URL, support contact, and an icon. Privacy and /contact exist. Docs can be a blog post or help page by publish date.

AI
```
Add a docs URL if missing (a blog post is enough). Icon that meets Claude’s listing size. Do not invent a new privacy policy.
```
Server is https://sails.app/mcp. Transport is Streamable HTTP. Not stdio.

Note
A directory listing is what feeds Suggested Connectors. 0 extra product work.

Note

Listing: name ≤100, tagline ≤55, description ≤2,000, slug (permanent after publish).

Write the Claude listing packet. Do not submit the slug until Ryan okays it. Voice: your AI assistant. Workshop not shop. 0 competitor names. Save sites/directories/data/claude-packet.json.

Team or Enterprise Claude org. Submitter has Directory permission (Owner by default).

Human

You need this before the portal appears in Claude.ai settings.
Test credentials: fully populated account. Write every URL, credential, and step.

Human, then AI

Same reviewer login as Shared if the server requires sign-in.
```
Write the reviewer sheet. Every link and step. Save sites/directories/data/claude-reviewer.md.
```

Every tool name ≤64. Every tool has a title and readOnlyHint or destructiveHint.

Reported 2026-08-28.

Claude rule, not ChatGPT’s three-hint rule. Same server change as shared/tool-hints for title + readOnly/destructive.