Docs
Connecting an assistant to Fraink.
This page is for the person doing the setup. If you are deciding whether Fraink is right for your brand, start with How it works. Everything here is the mechanics underneath.
Which assistants
Two ways in. Same permissions either way.
Assistants that support remote connectors sign in through Fraink’s standard OAuth flow. Anything else can use a personal access token. Both are tied to the person who connects and to what that person can already reach — nothing more.
| Assistant | How it connects | What you need |
|---|---|---|
| ChatGPT | Connector. Add Fraink under Connectors, then approve in your browser. | A ChatGPT plan that includes custom connectors. Codex uses the same connection. |
| Claude | Connector. Add Fraink under Connectors (not Plugins), then approve in your browser. | A Claude plan that includes custom connectors. |
| Grok | Personal access token. Issue it in Fraink, paste it into Grok’s secure field. | A Grok plan that lets it call an outside service with a token. |
| Gemini and others | Personal access token. Any assistant that can call an HTTPS API with a bearer token works the same way. | Depends on the assistant. Ask us; we will tell you honestly what we have tested. |
// plan requirements change often. we keep this table current; if something here is wrong, tell us.
Connector setup
ChatGPT and Claude
- Sign in to Fraink in your browser.
- In the assistant, add a new connector and paste this address:
https://fraink.com/mcp
- The assistant sends you back to Fraink. Approve the connection in the browser where you are already signed in.
- Done. The assistant can read your workflow and propose changes as you.
Do not open that address in a browser tab; it is for the assistant, and a “not authorized” message there is expected.
Token setup
Grok, Gemini, and others
- In Fraink, open your profile and issue a personal access token. It is shown once — copy it then.
- Paste it into the assistant’s secure credential field.
- Point the assistant at https://fraink.com/api/v1. It sends the token in the request, and Fraink checks the token’s owner and reach on every call.
Lost or leaked a token? Revoke it in your profile and issue a new one. The old one stops working immediately.
How the credential is handled
The details, for whoever asks.
- Storage
- Fraink never stores the raw token. Only a one-way hash is kept; the token itself is shown once at issue.
- Scope
- Nothing is frozen onto the credential. Every request re-checks the connecting person’s current roles, so a demotion or removal takes effect on the next call.
- Connector security
- Standard OAuth authorization-code flow with PKCE, dynamic client registration, and refresh tokens. Revocation is supported.
- Attribution
- Every change the assistant makes is recorded as the person who connected it. There is no separate agent identity in the audit trail.
- Public discovery
- Fraink publishes public facts for assistants at /llms.txt. Credentials never appear there.