Claude is the assistant we have tested the whole way through, from authorisation to answer. Once connected, it can answer questions about your actual spending, subscriptions, income and net worth — and it cannot change any of them.
An earmarkIQ account on Pro, with at least one bank connected, and an assistant that supports custom MCP connectors. Context is a Pro feature; it is not on the free tier or on Plus.
Custom MCP connectors live in Claude's settings on its paid plans. Look for Connectors, sometimes shown as Custom connectors. The exact location moves between releases and differs across web, desktop and mobile — Anthropic's own documentation is the authority if what you see does not match.
Paste the address below. It is the same address for every user — it is not a secret and not a token.
Claude will send you to earmarkIQ to sign in. You will see the five scopes listed and can grant them individually. Approve only what you want Claude to read.
If it is working, Claude will answer from your real figures rather than asking you for them.
This is the same for everyone. It is not a secret and it is not a token — authorisation happens in the next step, and the connection is bound to your account then.
Some clients cannot do the browser sign-in and ask for a bearer token instead. You can create one in earmarkIQ under Connected assistants. Manually created tokens expire after 90 days, and are shown once — if you lose one, revoke it and create another.
Five scopes, each granted and revoked on its own. You do not have to grant all of them, and you can change your mind later without reconnecting.
| Scope | What it lets the assistant read |
|---|---|
| Spending | Category totals and how they move over time — groceries, transport, eating out. |
| Transactions | Individual transactions: merchant, amount, date, category. |
| Balances & assets | Account balances, property values and the net worth figure built from them. |
| Income | Pay, when it lands, and how regular it is. |
| Subscriptions | Recurring payments, what they cost, and price rises. |
Transaction history goes back 12 months by default and is capped at 24 months. There is no setting that lifts the cap.
Context is read-only by construction rather than by policy. There is no tool in it that writes, moves or changes anything, because no such tool has been built — and a test fails the build if one is ever added. That is a stronger guarantee than a setting you could switch off by accident.
Read-only describes what Context can do to your data. It says nothing about what happens to the conversation once your data is inside the assistant. Whatever you ask Claude, ChatGPT or any other assistant is handled under that provider's privacy terms, not ours — and may be retained or reviewed according to their policy, not this one.
Grant the narrowest scopes that answer your question, and read the assistant provider's own terms before connecting anything. We would rather say that plainly than let read-only sound like a privacy guarantee it is not.
Every request an assistant makes is recorded and readable by you in the app: which tool ran, which scope it used, and when. It is not a summary or a sample — it is the list.
That matters more than any promise on a marketing page. A read-only claim you cannot audit is a claim you are taking on trust; a log you can open yourself is a claim you can check. If you connect an assistant and later wonder what it looked at, the answer is already written down.
Two routes, both immediate:
Revoking access to Context does not touch your bank connection or your earmarkIQ account. It only closes the door the assistant was reading through.