Tool · integration of AI agents with enterprise systems
MCP for 1C, Bitrix24 and email
2026 · Tool · Python, MCP
Portable MCP servers that give any AI client safe access to 1C, Bitrix24 and email
A set of MCP (Model Context Protocol) servers for Claude Code, OpenCode and our own 0-AGENT: 1C:Enterprise via OData V3 with automatic schema building from $metadata, the full Bitrix24 REST API, and email over IMAP/SMTP. Each one has domain-level RBAC (allow/deny by patterns), request rate limiting, retries and typed errors. To test the agent against a “real” 1C, a synthetic foreign trade and logistics test bench was built on the 1C 8.5 platform, with roles and test data.
The task
To connect an AI agent to accounting systems without rewriting the client for every 1C configuration and without the risk of the agent deleting or changing something it should not.
What’s inside
- 1C-MCP: 13 universal tools (entity list, metadata, select, read, count, create, update, soft and hard delete, posting and unposting, navigation, custom query); works with any configuration (Accounting, Trade Management, ERP)
- Two-level RBAC for 1C: 1C’s own roles plus a list of allowed and denied entities at the MCP level; hard delete is off by default
- Bitrix-MCP: calls to any REST method, batches of up to 50 commands, field introspection, CRM wrappers for deals, contacts, companies and leads; leaky-bucket rate limiting, exponential retry
- Email-MCP: 6 tools (inbox, read, search, send, folder list, attachments); provider-independent (Gmail, Yandex, Outlook)
- Handles the specifics of OData V3 in 1C: no $expand and no batch; 21 structured error codes are turned into typed exceptions
- Synthetic test bench: a freight forwarding configuration (4 enumerations, 7 catalogues, 4 documents with tabular sections), 3 roles and a service user, a verified permissions matrix
Machine translation — being proofread.