MCP for Abacus and Bexio
Prepare reports using data from Abacus and Bexio
Explain open items, check closing documents and prepare an actual-versus-plan comparison: we start with an agreed workflow for your finance team. Data from Abacus or Bexio and additional planning sources are included only within the approved scope. Results and calculation rules remain verifiable by the business team.
The way in: the official interfaces
For Bexio, we assess the REST API with OAuth 2.0; for Abacus, the interfaces available in your installation. Available data, permissions, versions and licensed modules determine the scope. Sign-in alone does not guarantee end-to-end permission mapping; this is tested for the selected functions.
The MCP server runs in your infrastructure or with a hosting provider you approve. The model location is agreed separately. Retrieved ERP data may be transmitted to the AI application and its model provider even when the server runs locally or in Switzerland.
Verify user permissions and client separation
Every request must map to a verified permission context. We test permitted and denied access to clients, data fields and functions. The pilot receives a limited data scope; production operation requires separate permission acceptance. Payroll and other particularly sensitive data remain excluded without explicit approval.
Possible tasks for your finance pilot
- Receivables and liquidity
- “Which open items over CHF 10,000 are older than 60 days, and who sent the last reminder?” instead of an open-items list that first has to be filtered.
- Preparing closings and VAT
- Have anomalies, missing receipts and unusual postings compiled before the quarterly or annual closing. The return itself remains with a human.
- Orders and quotes
- Query the status of quotes, orders and deliveries across customers, for example before a customer meeting.
- Clients of fiduciary firms
- Questions about the client base in plain language, with client separation through employees' permissions.
Data protection under the FADP
Source storage, model processing and logging are separate questions. We document which ERP data is transmitted to which recipients and where chat histories, logs or caches may be created. Hosting, model providers, retention and terms for data use are reviewed in advance and approved by you.
We provide the architecture information that the person responsible for data protection in your organisation needs for the assessment. This does not replace legal advice.
Limits we state up front
- No write access by default. No postings, no payments, no changes to master data without explicit approval.
- No circumvention of the Abacus or Bexio permission model.
- No submission of VAT returns or other filings by the AI.
- No additional storage or indexing without an agreed purpose, approved location and retention rules.
Frequently asked questions
Does our Abacus or Bexio data leave the company?
Retrieved content may leave the company if an external AI application or cloud model processes it. A local MCP server alone does not prevent this. A fully local solution also requires local model processing and appropriate settings for logs, telemetry and other services. We check the entire data flow before approval.
Does this work with Abacus in the cloud and on premises?
A connection may be possible in either case, provided your installation offers suitable interfaces and permissions. We check this with your IT team or hosting provider before planning the pilot. Version, licensed modules and available data determine the scope.
Does every employee see all the data?
Only the approved data scope may be accessible. User, role and client mapping is checked for the interfaces used and tested with both permitted and denied requests. Payroll data remains excluded without explicit approval.
Which AI platforms are supported?
Suitable MCP-enabled applications may connect to ChatGPT, Claude or local models. We review functions, authentication, data flow and provider terms for your specific selection and recommend an appropriate option without blanket promises about data residency or training use.
Let’s check whether it holds up for your systems.
We first clarify your use case, data sources, permissions and acceptance criteria. The feasibility review documents prerequisites and limits before a pilot is agreed.
Discuss your use case