IIkona (1c-ai-connector)
ИИкона (1c-ai-connector)
A 1C extension with an LLM connector, agent loop, RAG, error monitoring and an MCP server that exposes the infobase to external agents
Install
Download the .cfe from andromanpro/1c-ai-connector releases and give the user the steps: attach the extension, update the infobase, fill the AI models catalog, enter keys and an API user for MCP. Then connect the published MCP service HTTP address, or the 1c-ai-connector-mcp-proxy for stdio.This is third-party code. Review the repository files before installing.
What it does
The extension installs into a 1C configuration built on the Standard Subsystems Library and gives one programmatic interface to OpenAI-compatible APIs, Anthropic, Gemini, YandexGPT and GigaChat. On top of it run an agent loop where the model calls 1C tools, a knowledge base with embedding search in built-in storage or Qdrant, and error monitoring from the event log with LLM analysis and Telegram alerts. A built-in MCP HTTP service gives external agents the metadata tree, object structure, read-only queries and diagrams. The author ships a separate proxy for stdio clients.
Who it is for. For 1C developers embedding LLMs into standard and industry configurations and opening the infobase to agents.
Good fit when
- You need to call YandexGPT, GigaChat or other models from 1C code without writing a connector
- You want Claude or Cursor to read infobase metadata and data over MCP
- You need event log error analysis with Telegram alerts
Not a fit when
- The configuration is not built on the Standard Subsystems Library or the platform is older than 8.3.24
- You cannot install extensions into the live infobase or publish HTTP services
Example request
Connect to our infobase through the IIkona MCP server and show the structure of the sales document with its tabular sectionsLimitations
Requires platform 8.3.24+ and SSL 3.1.10+. Cloud models need their own keys and billing; OpenAI, Anthropic and Gemini APIs do not work from Russia without a VPN or proxy, while YandexGPT, GigaChat and local models do. The author notes that tool calling was tested live only on OpenAI-compatible models. The MCP server should be exposed externally only over HTTPS.
How to disable. Disable or remove the extension under Administration, Configuration extensions, and remove the MCP HTTP service publication.
MCP
- Transport
- http
- Authentication
- API key
Security check
- Installs into the live 1C infobase and exposes its data to external agents over HTTP
- Sends infobase data and error texts to cloud LLM providers
- Stores API keys and a Telegram bot token
README in short
The README lists the subsystems: a connector for five provider API formats, an agent loop, RAG, an MCP server with Basic Auth, rate limiting and audit, error monitoring and a diagram generator. Installation means attaching the .cfe from releases, updating the infobase and filling the models and experts catalogs. It shows BSL code examples for calling a model, platform and SSL requirements, and points to the guide, tests and proxy in separate author repositories. MIT licensed.
FAQ
Does it modify the main configuration?
No, it installs as a configuration extension and leaves the main configuration unchanged.
Where are API keys stored?
In 1C secure storage; they do not end up in configuration dumps.
Can it run offline?
Yes, with local models via Ollama or LM Studio, built-in knowledge base search and local mermaid rendering.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Anthropic's coding agent for the terminal, IDE and GitHub: understands your codebase, runs tasks and handles git