Frank, from whichever assistant you already use

Practice Settings → AI connections lets you connect an external AI client — Claude, ChatGPT, or anything that speaks MCP (Model Context Protocol) — directly to your practice. Once connected, you can read your pipeline, draft compliance content, and take real actions in the practice from your own chat client, not just the in-app assistant.

What it can do

A connected assistant can do anything your own role permits in the app — read leads and applications, check what's outstanding, draft a complaint response or a best-interests reasoning first-pass, advance a lead's stage, satisfy a compliance check, enable an automation template. Nothing more than your role already allows, and nothing you couldn't already do by clicking the equivalent button yourself.

Every action is still your action

A write action executes under your own live role's permissions — a demotion or removal takes effect on the very next call, not whenever the connection happens to refresh. Every action is logged to the same practice audit trail as anything done in the app or through Frank's own in-app chat, tagged so you can tell them apart. The two actions that record a real compliance attestation (best-interests reasoning, SOCA) get extra visibility in that log when executed this way.

Setting it up

1. Go to Practice Settings → AI connections.
2. Name the connection and choose which permissions to grant it — everything your role holds is checked by default; uncheck anything you don't want this particular assistant to have.
3. Confirm the consent notice — this genuinely grants the ability to take real actions in the practice, not just read data.
4. Copy the token and add it to your AI client as an MCP server, using the URL shown on the page.

Revoking access

Revoke your own connection any time from the same page. A principal or admin can see every member's active connections and revoke any of them — useful if someone leaves the practice or a token looks like it's been exposed.