API Key
An API key is the secret credential that ties every MCP tool call back to one specific user's own books, granting whatever that user's tools are able to do.
In short
A real credential, not a low-stakes connector. Anyone holding a valid key can call BalanceMCP's tools as that user — there's no narrower key limited to just reports, so a key should be revoked immediately if it's ever exposed, the same as a leaked password.
Also called: access key
An API key is the secret string that ties an MCP client's tool calls back to one specific user's books. It's what lets BalanceMCP know, on every single call, whose data is being requested or written to — the credential underneath the connection between an AI assistant and a real set of books.
It's worth treating an API key with exactly the seriousness a password deserves, not less because "it just connects an AI." Anyone holding a valid key can call every tool that user's account is able to call — import statements, post entries, run reports, reverse mistakes — as that user, indistinguishable from the account owner's own actions from the system's point of view.
A real, current limitation worth naming: there's no way today to issue a narrower key that, say, can only run reports but never post an entry. A key grants everything a user's tools are allowed to do, all at once — which means anyone trusted with a key should be someone trusted with the full set of nineteen tools, not a subset.
If a key is ever suspected of being exposed — shared accidentally, leaked in a log, pasted somewhere it shouldn't be — it should be revoked immediately, the same instinct as changing a compromised password, rather than waiting to see if anything goes wrong first.
What people get wrong
- Treating an API key as low-stakes because it "just connects an AI" — it grants everything that user's tools can do, the same as holding their password.
- Expecting a narrower key limited to reports only — that doesn't exist today; a key grants the full set of tools at once.
- Waiting to see if anything goes wrong before revoking a key suspected of exposure, instead of revoking it immediately.
Common questions
- Can I issue a key that only allows running reports, not posting entries?
- Not today — a key grants everything that user's tools are allowed to do, all at once. Anyone trusted with a key should be trusted with the full set of tools.
- What should I do if I think my API key was exposed?
- Revoke it immediately, the same as you would a compromised password — don't wait to see whether anything goes wrong first.
Machine-readable: /api/knowledge/concept:api-key