Era Context — giving my AI read-only sight of my money
I let an AI agent look at my bank and card transactions. Not by handing it a banking password, and not by scraping a screen — through a proper Open Banking connection, exposed to the assistant as tools. This is what that involves, what it costs, and why the free plan is the right answer for me.
What Era actually is
Era Context is two things bolted together:
- An Open Banking connection. It reaches my banks through Salt Edge, a regulated data aggregator. I authorise the connection once; my bank credentials are never handed to the AI, and Era does not store them.
- An MCP server. MCP — the Model Context Protocol — is the open standard that lets an AI client call tools on a server. Era exposes my financial data as tools rather than as another app with its own chat window.
The result is that the assistant I already use can ask “what have I spent this week?”, and get structured data back, under scopes I approve. The normal starting point is read-only.
Why I wanted it
I was already tracking the household money through spreadsheets and a manual bank export. The weakness was the export: periodic, hand-triggered, easy to skip. What I wanted was the boring part handled — the data arriving every day, shaped the same way every time, whether or not anyone remembers to do it.
How my setup runs
The point of my arrangement is that no LLM touches the data movement at all:
- Two scheduled jobs, 06:30 and 06:35 UTC. Each pulls accounts and transactions and stops.
- The data lands in a local SQLite database I own, refreshed daily.
- A plain Python script turns that into a weekly spending digest — spend by card, by category, with every transaction and description listed.
- Keys live in a local
.env, not in anything synced to the cloud. Read-only scopes. No write access, and nothing anywhere near trading.
A pull costs about 2 tool calls. That number matters shortly.
What the free plan actually gives
Read live off my own account rather than from the brochure:
| Free (Basic) | |
|---|---|
| Connected accounts | 2 |
| MCP tool calls | 100 (period unstated — see below) |
| Developer API requests | 500 |
| Transaction history | 21-day rolling window |
| Sync frequency | daily |
| Categories / rules | 2 custom categories, 0 pattern rules |
| Write access | yes, but I don’t use it |
Why free is enough for me — three reasons
1. I persist locally, so the history window is a one-off problem. The 21-day cap is a vendor-side limit on what they will show me. My database keeps everything it has ever fetched, permanently. Paying to look at history I am already storing myself would buy me nothing I don’t already have.
2. Two connections per login covers my setup. The free tier allows two linked accounts, which is my bank and my card. Where more are needed, a second login gives a second pair of slots, each with its own key and its own database — though the supported route to more connections is the paid tier, which allows 15.
3. The one thing I’d genuinely pay for isn’t on offer at any price. My ISA and SIPP sit with an investment platform, and that platform is not reachable this way. Open Banking is bank-and-card shaped — it was built for current accounts and payment cards. Investment platforms generally sit outside it, and the coverage that does exist is richest in the United States, which is Era’s home market. No plan tier fixes this. It is a limitation of the rail, not of the plan.
What the paid tiers add — and why I still don’t need them
| Free | Organize (~$84/yr) | |
|---|---|---|
| Connected accounts | 2 | 15 |
| MCP tool calls | 100 | 1,000 |
| Transaction history | 21 days | uncapped |
| Pattern rules | 0 | 15 |
| Sync frequency | daily | daily (no change) |
| Write access | yes | yes |
Running that against my three reasons: history → I persist locally. Connections → two is enough. Calls → see below. The pattern rules are the only genuinely attractive item, and they matter less to me because my categorisation is already built locally from my own taxonomy.
The MCP call quota, measured
This is the headline number on every paid tier, so it is worth measuring rather than guessing:
- One pull = 2 tool calls. Two pulls a day = ~4 calls a day.
- Paid tier allows 1,000 a day.
That is 0.4% used. My assessment (Alf): the call quota is not a reason to upgrade — for a deterministic daily pipeline like mine the headroom is irrelevant. It matters to people whose agent converses with the data all day; it does not matter to a cron job.
One caveat worth knowing if you are budgeting: the pricing page and the account’s own meter disagree on the free tier’s call allowance (the page says 250 per day; the account meter says 100, with no period stated). The meter is what gets enforced, and neither figure is anywhere near a constraint at four calls a day.
The honest limitations
- Joint accounts lose attribution. On a joint account or a shared card, the feed tells me which card was used, never which person. There is no cardholder field on any surface of the API. For a household, this is the real missing feature — and no tier provides it.
- It is bank-and-card shaped. No investment platforms, as above.
- Aggregator categories are weak. Utility bills, insurance and council tax all arrived in one “cash and misc” bucket. I replaced them with a taxonomy learned from my own export.
- Daily sync on the cheap tiers. Fine for a daily digest, wrong for anything intraday.
If you want to try it
- Sign up free and connect one account.
- Ask a read-only question first. Confirm what the assistant can and cannot do before granting anything more.
- Keep keys out of anything that syncs to the cloud. Prefer OAuth over long-lived API keys where your client supports it — and remember revoking access stops future reads but does not unwrite what was already shared.
- Leave write access off until you have a reason for it. An assistant that can read your money is useful; one that can edit it is a risk you should choose deliberately.
My assessment (Alf)
Era Context is the cleanest way I have found to give an assistant financial context without giving it credentials. The free tier is a genuinely usable product rather than a demo, on one condition: you persist the data yourself, which I do. The paid tiers are reasonable value for people who want the vendor to hold their history, or who run many accounts — neither describes me, so I stay on free. The gap I would actually pay to close is the missing investment-platform feed, and no plan tier closes it.
Read-only analysis, not financial advice.