Developers
The same data, the same permissions, over an API.
Everything the product does, it does through procedures that are catalogued, typed and permission-checked. The API is not a slice of that surface reimplemented for outsiders — it is the surface.
- 162typed MCP tools
Model Context Protocol
Bring your own AI. Point it at your own books.
aicroo is an MCP server, so the assistant your team already uses can ask real questions of real books — what is overdue, which supplier drifted, what the VAT position looks like — and answer from your ledger rather than from a guess. It is the same tool surface our own assistant runs on, exposed to yours.
Typed tools
Each tool declares its arguments and its result, so a model is schema-checked at the call rather than trusted afterwards.
A catalogued procedure surface
Beyond the curated tools, the full catalogue of business procedures is discoverable and callable by name — so a subsystem with no dedicated tool is still reachable.
It runs as a principal
An MCP session carries a real identity with real permissions. There is no privileged read-everything mode to connect with.
Refusals are ordinary
Asking for something the key cannot see returns a refusal, not a partial answer — and the refusal is recorded.
Connect a client
One URL and a key.
The server speaks MCP over HTTP, which is what remote-connector support in the current generation of assistants expects. Anything that speaks it can connect — Claude and ChatGPT among them, as well as your own agent built on any SDK. Nothing is installed and nothing runs on your machine.
Endpoint
https://aicroo.cloud/rest/mcp
Authenticate with either
- X-API-Key: <your key>
- Authorization: Bearer <session token>
Issue the key to the narrowest role that does the job. What the client can read and write is decided by that role in the database — not by the client, and not by how the question was phrased.
Issue a key
Create an API key for your company, and give it roles the same way you would give them to a person.
Add the connector
Paste the URL into your client’s custom / remote MCP connector settings and give it the key as a header.
Ask it something
The client discovers the tools by itself. “Which invoices are overdue by more than thirty days, and who owes the most?” is a reasonable first question.
Let it do the work
Reads are only half of it. The same surface creates documents, reconciles bank lines and files a settlement — under the key’s permissions, with the irreversible steps gated behind a preview.
It is not a read-only window
The tools cover the ledger, sales and purchases, payments, bank reconciliation, Spanish tax settlements, fixed assets, absences, approvals, signatures, loans and stock. Posting, voiding, filing and disposing are rehearsed first: the call returns a preview and a one-time token, and committing means calling again with that exact token.
REST
Keys that behave like people.
An API key is issued to a company, given roles through the same chain a person is, and revoked the same way. What it can read and write is decided by those roles in the database, not by the endpoint it happens to call.
Authenticate
Present your key and exchange it for a short-lived session, or send a session token directly.
Scope it
Assign the key the narrowest role that does the job.
Call it
Read and write through the same procedures the product itself uses.
Revoke it
Removing the key removes the access, everywhere, immediately.
In and out
A workflow is an integration you did not have to host.
Before you write a service against the API, check whether a flow already does it. A flow can be started by a signed URL from outside, can call outward to another system, and can run your own code in a sandbox — with retries, run history and a rehearsal mode you would otherwise have to build.
Good to know
Two things worth knowing up front.
Everything writes with a key
The one exception is the demo request form, and it is well guarded: it validates everything server-side, rate-limits at the edge and drops what looks automated.
A catalogue that keeps itself current
The procedure catalogue is derived from the schema, so it always matches what the database exposes — and grows with it.