AI Finansist: MCP server for 1C
mcp_1c (AI Финансист)
An MCP server, a 1C OData client and a local language model bundled with a Gradio chat for questions about 1C data and uploaded PDFs
High risk
We rate an entry high when the tool writes to external systems, handles money, production databases or secrets, or runs arbitrary commands. The CLI installs it only with your consent.
Why this level
- The OData client can create, update and delete 1C database records, not just read them
- The 1C login and password are stored in .env in plain text
Install
Manual install
docker compose up --buildBefore starting, fill .env with the 1C OData address, login and password, and bring up vLLM separately.
This is third-party code. Review the repository files before installing.
What it does
The project brings up a four-part stack via docker compose: a vLLM server running the Salesforce xLAM model for tool-call parsing, an MCP server with 1C tools built on the OData protocol, an orchestrator that ties the model's output to MCP calls, and a Gradio web chat. The OData client can read, create, update and delete data in a 1C database through the standard standard.odata interface. The chat also accepts PDF uploads, whose text the pdf_parser module extracts and feeds into the conversation alongside the 1C data.
Who it is for. For 1C developers who want their own local chat with access to 1C data over OData, without cloud LLM providers.
Good fit when
- You want a fully local stack: your own model, your own MCP server, your own chat, no external APIs
- Your 1C data is reachable through the standard standard.odata interface
- You want to combine questions about PDF documents and 1C data in one chat
Not a fit when
- You lack the hardware to run the local Salesforce xLAM-2-32b model through vLLM
- You need an MCP server for Claude Code or Cursor, not a separate web chat
- Your 1C database does not expose OData, or exposes it on a non-standard path
Example request
Show the stock balance from the 1C catalog and compare it with the table from the uploaded PDF price listLimitations
No updates for over a year, and the README does not document the MCP tool architecture; documentation is limited to project layout and environment variables. It needs a separately running vLLM server with the Salesforce xLAM-2-32b model, a real GPU load. The 1C database login and password go into .env in plain text. The MCP transport and auth are not documented in the README; check mcp_server.py source for exact settings.
How to disable. Stop the stack with docker compose down and remove the project's containers.
MCP
- Transport
- http
- Authentication
- not required
| Environment variables | |
|---|---|
| MCP_1C_BASE required | Address of the 1C OData database, e.g. http://host/TEST19/odata/standard.odata |
| ONEC_USERNAME required | 1C user login for the OData connection |
| ONEC_PASSWORD required, secret | 1C user password, stored in .env in plain text |
Security check
- The OData client can create, update and delete 1C database records, not just read them
- The 1C login and password are stored in .env in plain text
README in short
The README gives the project layout: docker-compose.yml, gradio_app.py for the chat, orchestrator.py as the agent between the LLM and MCP, mcp_server.py, odata_client.py and pdf_parser.py. It describes running vLLM with the Salesforce xLAM-2-32b model on port 8000, a sample .env with MCP and 1C OData addresses, login and password, and the final docker compose up --build launch with the Gradio UI and MCP server addresses.
FAQ
Does it need access to cloud LLMs?
No, the README describes running the local Salesforce xLAM-2-32b model through vLLM on your own hardware.
Can this be used as a regular MCP server for Claude Code?
The README does not describe that; the project is built as a self-contained Gradio web chat, not an MCP server for an external client.
Related
A skills library that gives coding agents a development process: brainstorming, planning, TDD, subagents and code review
Skills for real engineers by Matt Pocock
Skills For Real Engineers
Small composable skills for engineering with agents: plan grilling, TDD, bug diagnosis, code review and architecture
GitHub toolkit for spec-driven development: the specify CLI adds agent commands and skills to a project, from principles to implementation
Reference MCP servers
Model Context Protocol servers
Official reference MCP servers: Filesystem, Fetch, Git, Memory, Sequential Thinking, Time and Everything