One paperwork path, three doors in.
A growing firm still lives on paper. An invoice, a delivery note, a cash receipt. Someone looks at it. Then it is written down. Other document types can come later. This is not a full accounting product.
- Starting point
- Agent
- Knowledge
- Decision
- Tool
- Approval
- Result
A pragmatic workshop, not an ERP
Told as the path of a growing workshop: incoming invoices, the cash register, one warehouse with categories, people — later performance and bonuses. Records stay in tables you already keep. Not a named ERP, and not live paperwork at SM Servis.
Enough structure to run the workshop as it grows: incoming invoices, a real cash register, one warehouse with categories, a staff list. Performance and bonuses come later. Records live in your SQL, REST or tools — not in a branded ledger we do not sell.
Incoming invoices
Supplier, number, dates, amounts, VAT, due date, lines if present. After approval, insert into your invoices table or a CSV.
Cash register
Cash receipts and related cash documents — enough to keep a workshop till, not a full statutory ledger.
One warehouse, several categories
A single stock location with categories (parts, fluids, and so on). Delivery notes update stock via SQL or MCP — not a multi-site WMS.
Employee list
Who works in the shop. A table you already keep. Company-wide directory login stays on its own track.
Performance and bonuses
Measures from work already recorded, then a bonus calculation a person still approves. Later — not live at SM Servis.
How it starts — same path, three ways
Forward an email with files
A message in the connected mailbox starts the work. Attachments stay files — they are not poured into the request.
See emailUpload into Files
Photos and PDFs can be uploaded, including a drop link. They stay files, separate from searchable knowledge. Starting this path automatically when a file lands is still being finished.
See FilesAttach files to a chat message
Website chat is live. Attaching a file so this paperwork path starts is still being finished.
See website chatTypes in the first cut
Then the work
Detect the type first
Invoice, delivery note or cash receipt first; other types later. A dedicated reader for that classification is still being finished.
See automated process stepsThen take the matching path
Once the type is known, the work already continues down the matching branch.
See automated process stepsFill the fields for that type
The numbers and names that type needs can be filled in. There is no branded invoice reader in the catalog.
See automated process stepsA person must approve
Work already waits for a person. Staff get a mail or a button in the channel they already use.
See control and governanceDocument on one half, fields on the other
A split view of the paper and the extracted fields, changing the type and reading again, is still being finished.
See control and governanceWrite after approval
Then write into your records, store a file, or call a tool you already have. Only these destinations are claimed.
See data connectionsWhere approved data can go
A table you already keep
After a person says yes, the numbers can be written into tables you already use.
A table you already keepThe systems you already call
The same records can go out over the connections you already have — not an open shell on a laptop.
The systems you already callA file you can open
Approved data can also land as a file in Files — a spreadsheet you can download, not a hidden export step.
A file you can openA write through a tool
If you already have a governed tool that writes, the same path can call it after approval.
A write through a toolMatch the fields
Names on the paper become names in your table — you say which field is which.
Match the fieldsHow the workshop grows
First the paper is read and a person says yes. Then invoices land in your records. Then the till and the stock. People and bonuses come last.
What already works
Mail can start the work. Files can hold the paper. A person must approve. Then the numbers can be written into records you already keep, or into a file you can open.
What we are still finishing
A clearer screen with the paper on one side and the fields on the other, starting automatically when a file arrives, and attaching a file from chat.
Read the paper, then wait for a person
Mail comes in. Type and fields are proposed. A person says yes or no. A clearer split view of the paper and the fields is still being finished.
Approved invoices into your records
After approval, write into a table you already keep, or into a file you can open. Incoming invoices for a growing workshop — the review screen is still being finished.
Cash register and warehouse
The same path, more papers: cash documents, one warehouse with categories. Still your database — not a named ERP.
People, then bonuses
A staff list, then performance from recorded work, then a bonus a person still approves. Later for the growing workshop.
Related pieces
Do not read this as SM Servis already processing invoices in production. There is no branded accounting app in the catalog.
Small and mid-size firms
Mail that becomes a reply, paperwork that waits for a person, invoices, and a company calendar where appointments are written — visits, meetings, bookings — without a double booking.
Connect a mailbox (classic IMAP or Microsoft 365). A new message starts a process, attachments are stored, and they do not belong in the model’s prompt. The same kind of process can start in Teams or on WhatsApp.
Datastore
Searchable text lives in the database. Photos and PDFs are separate — browse them in admin and use drop-off links.
Automated process steps
From starting points and agents to approvals, actions and notifications. The company calendar is one of those steps — an appointment in the calendar, not an engineering shortcut.
Control and governance
AI can be highly capable, but in a business it must also follow clear rules. Before an action runs, MCP Fabric can check the organization, identity, role, available services, licensing, policies and capacity. Sensitive actions can require human approval. This gives you control not only over what AI can access, but over what it is actually allowed to do.
References
These websites show concrete MCP Fabric use cases in practice — from customer conversations and business knowledge to approvals and operational processes. The capabilities shown on each site reflect what that organization actually has enabled.
Ready to move beyond the chatbot?
Build AI that understands your business, works with your knowledge, follows your processes and can actually finish the job. You decide which actions are allowed, when a person must step in and how much AI usage your organization can consume.