Troubleshooting

Check the URL is exactly https://cortex.nativz.io/api/mcp/user, no trailing slash, no spaces. If your client caches connectors, remove Cortex and re-add it. Corporate VPNs and proxies that strip streaming responses can also block MCP; try another network to rule that out.

Your browser is probably blocking the popup. Allow popups for your client, or copy the auth URL it prints into a browser manually. In Claude Code, run /mcp, pick cortex, and choose Authenticate to retrigger it.

Most clients need the connector enabled per conversation: check the tools menu in your chat. In ChatGPT, the connector only appears after enabling Developer mode under Settings → Connectors. Restarting the client after first setup fixes most of the rest.

MCP sessions are scoped to the clients your Cortex account can access. If a brand is missing, it’s a permissions question, not a bug: ask your Nativz contact to add you to that client.

Sessions expire and refresh tokens can be revoked when your password changes. Just approve the sign-in again; your connector config doesn’t need to change.

The header must be exactly Authorization: Bearer ntvz_.... Watch for a missing Bearer prefix, smart quotes from a doc, or trailing whitespace pasted with the key. Keys are shown once at creation; if it’s lost, revoke it and create a new one in Settings → API keys.

API access is enabled per organization by your Nativz team. If key creation shows API access is disabled for your organization, contact your account manager to turn it on.

Two causes. Either your key is missing the resource scope for that endpoint (each key carries scopes like clients, search, and posts), or the resource belongs to a client outside your organization. Create a key with the scope you need in Settings → API keys, and only act on clients returned by /api/v1/clients.

Usually the client_id is wrong or belongs to an org your key can’t see. List /api/v1/clients first and use IDs from that response only.

Keys are limited to 30 requests per minute over a sliding window. Read the Retry-After header (seconds) and back off for that long; hammering it immediately just extends the window. Batch reads where you can.

POST /api/v1/search only queues the search in pending status; a Nativz teammate runs the research from the Cortex app. There is no v1 endpoint that returns results, so the status will not flip on the API. Review completed research in the Cortex app.

That’s intentional. API keys must never ship to a browser; anyone could read them. Call Cortex from your server and have your page talk to your server.

Check the post’s status first: it may still be awaiting review. Times are interpreted in the client’s timezone, so a "wrong time" is usually a timezone mismatch in scheduled_at.