Get VRDex listed in the ChatGPT app directory and the Claude Connectors Directory

[BASIC-BIT]

https://vrdex.net/mcp is live and connects. Both assistant platforms now accept directory submissions. This issue tracks what it takes to actually get listed on each, and what is already done.

Related: #268 (discoverability on our own site), blocked by #269 (privacy policy and terms).

The two targets are not the same shape

ChatGPT app directory — open for submissions. No organization tier gate.

Claude Connectors Directory — open, but the submission portal lives inside Claude.ai organization admin settings, so it requires a Team or Enterprise organization, and only Owners or Primary owners can submit. That is a subscription cost incurred purely to be allowed to submit.

Both self-serve paths already work today and need none of this: a Claude custom connector is Customize → Connectors → + → paste the URL, on any plan including Free. ChatGPT dev mode is Settings → Security and login → Developer mode. Directory listing is about distribution, not capability.

Already satisfied

• Hosted Streamable HTTP MCP at https://vrdex.net/mcp, publicly reachable — Claude connects from Anthropic's cloud rather than the user's device, so this matters
• Anonymous public reads, advertised as noauth in _meta.securitySchemes, so a reviewer can exercise read tools before authenticating anything
• OAuth with Dynamic Client Registration and public-client CIMD — OpenAI recommends CIMD and supports DCR; Claude treats OAuth client ID/secret as optional
• search and fetch aliases in the exact required shape — search results carry id/title/url, fetch carries id/title/text/url plus optional metadata
• Accurate tool annotations on every tool — readOnlyHint, destructiveHint, idempotentHint, openWorldHint. Worth calling out because "inaccurate or missing tool annotations" is a named ChatGPT rejection reason, and ours are deliberate: destructiveHint: true on vrdex_profile_update, false on vrdex_profile_submit
• Specific, non-misleading tool descriptions, approved 2026-08-11 and 2026-08-12
• Public documentation at docs.vrdex.net
• Passes the originality bar — submissions must offer "functionality or workflows that are not natively supported by the products' built-in capabilities". A dataset with authenticated writes is not a rephrasing of something the assistant already does.

Missing

Blocking both (see #269):

• Public privacy policy URL — immediate rejection on Claude without one
• Terms of service
• Support contact reachable by end users, matching the listing

Reviewer access, needed by both:

• A demo account with sample data, populated enough to exercise every tool end to end — including a claimed profile the reviewer can edit, since the write tools are the interesting half. ChatGPT rejects authenticated servers submitted without working test credentials. Claude asks you to attest you have run every tool yourself via MCP Inspector or as a custom connector.
• Written setup instructions covering every link, credential and step

ChatGPT-specific:

• Verified developer or business identity on the OpenAI Platform — reviewers match it against the listing's name, website, support contact, privacy policy and terms
• Directory metadata: listing copy, starter prompts, test cases, country availability, policy attestations
• Icon, and screenshots at the required dimensions if included

Claude-specific:

• A Team or Enterprise Claude organization, to reach the submission portal at all
• Icon and carousel screenshots
• Seven policy acknowledgments: directory guidelines, first-party API usage, financial transactions, AI media generation, prompt injection, conversation data collection, public documentation

Verification still outstanding:

• docs/developers/mcp-client-compatibility.md still records the ChatGPT Apps/Connectors UI row as pending. What passed is the OpenAI Responses API (gpt-5.6-luna, data-backed search and fetch), which is API integration evidence, not product-surface evidence. Walk the dev mode path end to end and update the row.
• Confirm whether the deep-research connector path requires an SSE endpoint specifically; OpenAI's MCP documentation references /sse/, while developer mode accepts streaming HTTP. Developer mode is unaffected either way.
• Exercise the hosted OAuth write path from a real client. Everything to date is in-process.

Suggested sequencing

ChatGPT first. No organization tier gate, submissions are open, and the annotation and description hygiene that commonly trips submissions is already done. The remaining work is #269 plus a demo account and listing metadata.

Claude directory second, once the Team or Enterprise question is decided. Worth noting the self-serve custom connector already gives Claude users the full experience on any plan, so directory listing there buys discovery rather than capability — which makes it a marketing decision with a subscription attached, not a prerequisite.

Note on the ChatGPT consumer story

Developer mode is self-serve but sits behind a settings toggle that OpenAI wraps in an explicit prompt-injection warning. It is not the "paste a URL and go" experience #268 is aiming at. On ChatGPT, the actual consumer path is a directory-listed app, which is why this issue matters more there than it does for Claude.

[BASIC-BIT]

[AGENT]

The existing checklist item to exercise the hosted OAuth write path from a real client is now satisfied for the Codex desktop operator surface. On 2026-08-31, a project-scoped production OAuth session completed one community profile submission and a follow-up profile update, including public readback. No profile identifiers are retained in the durable evidence.

A fresh production protocol smoke on 2026-09-01 also passed anonymous reads, data-backed search and fetch, OAuth metadata, DCR, CIMD, and discovery of mcp:read, profile:read, mcp:write, assets:write, events:write, profile:write, and profile:contribute.

PR #283 records the client proof separately from the still-pending ChatGPT Apps and Connectors product-surface row: #283

This does not complete the directory-listing issue. Privacy and terms, reviewer access, listing metadata, and actual ChatGPT Apps and Claude directory submission evidence remain open.