ChatGPT

ChatGPT connects over OAuth in developer mode, from OpenAI's cloud, so the server must be public over HTTPS.

Add the app

1. In ChatGPT open Settings, Apps and connectors, Advanced, and turn on Developer mode.
2. Create a connector. Name Nexara, MCP server URL https://dev.nexara.ac/mcp, authentication OAuth.
3. ChatGPT registers itself (POST /oauth/register) with the callback https://chatgpt.com/connector_platform_oauth_redirect and opens the Nexara consent page. Pick Spaces and a ceiling (up to confidential) and approve.

To pin the connector to one workspace, use https://dev.nexara.ac/w/<workspace>/mcp as the URL; the token is then bound to that URL only.

Status in 0.1.0

• Works through dynamic client registration and PKCE. The authorization response carries iss (RFC 9207), and client_id_metadata_document_supported is advertised.
• Deep Research and company knowledge work through the search and fetch tools, which follow OpenAI's compatibility schema. search takes one query string and returns {results: [{id, title, url}]}, one result per Page. fetch takes one id from those results and returns {id, title, text, url, metadata}. Both are read-only, declare an outputSchema, and obey the same grants, ceiling, redaction and audit as context_search and context_get.
• Every url is the absolute web link to the Page (https://dev.nexara.ac/n/<id>), so ChatGPT can cite it and you can open it in the Nexara app.
• In developer-mode chat ChatGPT can also call every context_* tool directly. Per-tool securitySchemes are not shipped yet.
• OAuth access tokens work on MCP only; /api/v1 refuses them. Use an agent key for REST.
• Tool names are underscore-only (context_bundle), which is what OpenAI function names require.

Verify

1. Ask ChatGPT to "call context_whoami on Nexara". The connector shows up on the Agents page with its grants.
2. Deep Research: start a Deep Research run with the Nexara connector selected and ask about something you know is in the vault. ChatGPT calls search, then fetch on the hits. The answer cites Nexara Pages, and each citation opens the Page at /n/<id>. The reads appear on the Audit page as search and node.read events for the connector.
3. Redaction check: ask about a Page that has a block above the connector's ceiling. The fetched text shows [REDACTED: <level>, ask owner] in its place.