Skip to content

ConsentAlly works with AI assistants. A merchant can have their assistant set up, sign in to or check their ConsentAlly account from the chat: the assistant needs only the merchant's email address, and the merchant approves it on one ConsentAlly page with a code. Nothing is created or connected before that approval. How it works: https://consentally.com/auth.md

Docs pages

How an assistant signs a person up

An AI assistant signs a person up for ConsentAlly without filling in any form. It asks for their email address, gives them one ConsentAlly link and a 6-digit code, and the person approves on that page; their account, and the website if the assistant named one, is created only when they approve. An assistant that takes MCP connectors does the same through OAuth, with no code to type.

Two ways to connect

As an MCP connector

Claude, ChatGPT, Cursor, VS Code and other apps that take MCP connectors add one address:

Text
https://app.consentally.com/api/mcp

The app opens ConsentAlly's page in the browser. The person signs in, or creates their account there, and chooses the organisation, what the assistant may do and which websites. A person with no organisation yet gets one when they approve, and is sent straight back to the app.

Each app's steps are in Connect an AI assistant.

Through auth.md

Any assistant that can make HTTP requests follows auth.md, which has every request and answer. In short:

  1. Ask for the person's email address, unless you already know it. Send the website too if they named one; do not ask for it.
  2. POST https://app.consentally.com/agent/identity with {"type": "service_auth", "login_hint": "<their email>", "website": "<their website>", "client_name": "<your name>"}. website and client_name are optional; for a Shopify store, send its myshopify.com address, and installing the ConsentAlly app from Shopify afterwards can take that website over.
  3. Send the person the answer's human_instruction word for word: one sentence with the link and the code.
  4. The person opens the link, signs in or creates their account (verifying their email with ConsentAlly), types the code, chooses what you may do and approves.
  5. Poll the token endpoint (POST https://app.consentally.com/oauth2/token, with the claim grant auth.md shows) every 5 seconds. The answer is an access token, valid 1 hour, and an identity assertion, valid 90 days, which gets new access tokens without the person.
  6. Call next_steps and follow the next of every answer.
The code works for 10 minutes; when it runs out, the claim endpoint gives a fresh link and code for the same email address.

What not to do

  • Do not fill in ConsentAlly's sign-up or sign-in forms in a browser for the person, and do not ask them to create an account first. The link does both.
  • Never show the person a token, never ask them to paste one, and never put one in a web address. ConsentAlly refuses a token sent in the query string as leaked.
  • Do not guess the person's email address. The approval page only lets the person who holds that address, signed in and verified, approve.

Someone who already has an account

The same link works. The person signs in, picks one of their organisations and its websites, and may let the assistant change things only where they are an organisation admin; anyone else can grant read only.

What ConsentAlly itself does not do is on Limitations.