Frequently asked questions

How do I connect an application?

From the dashboard's Connectors page, click Connect on the application's card. For OAuth apps (GoTo Webinar, HubSpot, Sellsy…), the provider's own sign-in window opens: you approve, you are sent back, done. For API-key apps (Ringover, Aircall, Mailchimp…), a small form opens instead — the card tells you exactly where to create the key in that application.

Connections are per user: each teammate connects their own accounts, and their AI acts with their own permissions — never someone else's.

How do I connect a self-hosted application (WordPress, Dolibarr, Magento)?

Same place, but the form also asks for your instance's URL, because it is yours rather than a fixed vendor domain. The instance must be reachable over HTTPS and expose its REST API.

Minimum versions: WordPress 5.6 or later (application passwords), Dolibarr 11.0 or later, and Magento 2 with an integration access token (System → Extensions → Integrations).

The connection is refused — key rejected, or endpoint not exposed?

“That key was rejected” means the application itself refused the credential: re-copy it, and check it has the required permissions.

“This instance doesn't expose the REST endpoint Proxoria needs” is different: your credentials may be perfectly fine, but the instance cannot answer — its REST API is disabled (a security plugin on WordPress, the API module on Dolibarr) or its version is older than the minimum. Fix it on the instance, then try again.

How do I connect Claude, ChatGPT or Mistral to Proxoria?

Add Proxoria as a custom connector in your AI client — in Claude: Settings → Connectors → Add custom connector, with your Proxoria MCP URL. The first use opens a Proxoria consent screen where you sign in and choose which of your connected applications the AI may reach — or tick “applications I connect later” to include future ones automatically.

ChatGPT and Mistral follow the same flow: they all speak the Model Context Protocol.

My AI does not see an application I just connected

Three steps, in order. First, start a new conversation: Claude fixes its tool list when a conversation starts, so an ongoing chat will not pick up new tools.

Still missing? In your AI client's settings, disconnect the Proxoria connector, reconnect it, and start a new conversation again.

Finally, if you had selected a fixed list of applications on the consent screen (rather than “applications I connect later”), the new app is deliberately excluded from that authorization — reconnecting reopens the consent screen so you can include it.

Does connecting many applications degrade my AI?

Each connected application adds its tools to what your AI sees. Beyond roughly 30–40 tools, assistants become less precise at picking the right one. If you connect many applications, use the consent screen to scope the AI to the ones you actually use from it.

Can the AI modify or publish things in my applications?

Write tools exist, but two guardrails apply. In confirmation mode (the default), every write action waits for your approval on the Activity page before anything executes.

And whatever the AI creates is born unpublished — WordPress posts and pages as drafts, Magento products and CMS pages disabled, Intercom translations as drafts. Publishing stays a human action, in the application itself.

How are my credentials stored?

Encrypted at rest (AES-256-GCM, with the key held outside the database), never logged, and never included in error messages. Disconnecting a connector deletes its credentials. Every tool call your AI makes lands in the audit log, visible on the Activity page.

What is my Proxoria API key, and what if it leaks?

It is the personal key (pxa_…) that identifies your environment — your connectors, nobody else's. It is shown exactly once when generated. If it leaks, rotate it from the dashboard: the old key stops working immediately.