An AI Agent for 1C: What It Can Do and How to Connect It
How to connect an AI agent to 1C, the business software widely used in Russia and the CIS: answers about stock, orders and documents, plain-language reports, incoming documents. Integration options and safety rules.
1C is the accounting and ERP platform most businesses in Russia and the CIS run on: stock, orders, customers, documents, money. Yet getting an answer to a simple question out of it often needs someone who knows where everything is. An AI agent connected to 1C answers those questions itself: "how much is in stock", "what has not shipped", "who has not paid".
What an AI agent can do with 1C
Answer questions in plain language
"Which orders from this week have not shipped?", "How much of product X is left in the Krasnodar warehouse?", "Which customers are more than a month overdue?". The agent turns the question into a data request, gets the answer and explains it.
Prepare summaries
Every morning or on request: sales, stock below minimum, overdue payments.
Process incoming documents
An invoice or delivery note arrives by email. The agent recognises it, matches it to the order and prepares a draft for posting. A person posts it.
Help staff
An internal bot: "how do I process a return", "where is the reconciliation statement", with hints based on your rules.
How to connect an AI agent to 1C
- 01Questionstaff ask
- 02Agentworks out what is needed
- 03MCP serverthe agent's tools
- 041CHTTP service or OData
- 05Answerin plain language
There are several routes; the choice depends on the configuration and version:
| Option | What it is | When it fits |
|---|---|---|
| The standard OData REST interface | Built-in HTTP access to data | When you need to read catalogues and documents |
| HTTP services | Custom methods a 1C developer builds for the task | When you need specific ready answers and control |
| Exports | Regular export of the needed data to a database or spreadsheet | When direct 1C access is not allowed |
| An MCP server on top of one of these | A set of tools for the agent | When one or more AI agents work with 1C |
The most sensible setup is usually this: a 1C developer builds a few HTTP services for the needed questions, and an MCP server on top of them is what the agent uses.
Safety
- Read-only at first. The agent answers and drafts, but posts nothing itself.
- A dedicated 1C user with minimal permissions.
- Limit the data. An agent answering about stock does not need payroll.
- A request log. You can see what the agent asked and when.
- Where the model runs. If data cannot be sent to external services, you need a self-hosted model or anonymisation.
Where to start
- List the 10-20 questions about 1C staff ask most often.
- Decide which data they need and who may see it.
- Choose the access method with your 1C specialist.
- Build an agent that answers only those questions.
- Expand the list as the agent proves accurate.
FAQ
Is there a ready-made AI agent for 1C?
Ready solutions are appearing, but every 1C configuration is different. You will almost always need to tune it to your data and questions.
Can AI post documents in 1C by itself?
Technically yes, but I do not recommend it at first. Let the agent prepare drafts and a person post them. Write access comes later and selectively.
Do I need a 1C developer?
For a reliable integration, yes, at least to expose the data and build the services. The rest is done on the agent side.
Is it safe to give 1C data to an AI model?
It depends on the data and the model. For sensitive data, use self-hosted models, minimal permissions and anonymisation.