gf-data-1c
A read-only Rust MCP server for 1C:Enterprise data that warns about silent query traps, and lets the platform build the query text instead of the model
Low risk
We rate an entry low when it mostly gives the agent instructions and reference material.
Why this level
- Read-only by design; a change request is rejected before it reaches the base
Install
Manual install
gfdata.exe -install "C:\путь\к\базе" -db-user <пользователь> -db-password <пароль>Installs the extension into a file or server base; the binary carries the extension inside it.
This is third-party code. Review the repository files before installing.
What it does
The server installs an extension into a 1C base and reads data through it: balances, turnovers, information register slices, accounting account totals, catalogs and record counts. It is built around four fixes for specific cases of the model getting confidently wrong answers: it warns about silent query traps like a reference compared to a string literal or a balance taken at day-start with no time; it only runs query text that was built by its own query_build tool in the same session; it enriches a platform refusal with a hint listing available fields and replacements for language constructs that do not exist; and it serves syntax and type-member reference straight from the installed release's own documentation rather than the model's memory. A refusal from a tool never looks like an empty result, and the base is always named explicitly. A call without one is rejected by the schema.
Who it is for. For 1C developers and analysts worried that an agent might produce a plausible but wrong report from base data.
Good fit when
- You need read-only access to balances, turnovers or accounting totals through an agent
- You need protection against silent query mistakes like a reference compared to a string
- You need precise query-language and object reference for the exact installed release
Not a fit when
- You need to write data: the server only reads by design and rejects change queries before touching the base
- The configuration is not Russian-language: query-notation translation targets Russian name prefixes
Example request
Show the stock of Цемент М400 at the main warehouse as of today and reconcile it with the accounting totalsLimitations
Russian-language configurations only, by design. Read-only; changes are rejected before touching the base. Building the server with the extension embedded needs proprietary 1C object-build tooling that a third-party developer does not have; without it the extension is installed from the prebuilt dist/GTData.cfe file. A base response larger than 128 MiB is rejected with a clear refusal.
How to disable. Remove the base from the server's registry with bases action=remove and delete the GTData extension from the base's configuration.
MCP
- Transport
- stdio, http
- Authentication
- API key
| Environment variables | |
|---|---|
| TOKEN secret | Bearer token for the network HTTP mode; empty means no check. |
| DATA_DIR | Directory for the base registry and log in container mode. |
Security check
- Read-only by design; a change request is rejected before it reaches the base
README in short
The README explains the product philosophy in detail: four mechanisms against specific cases where an agent gave a plausible but wrong answer, a table of thirteen tools, extension install via gfdata.exe -install or a prebuilt .cfe, registering the base in a registry with DPAPI password protection on Windows, running in a container with explicit warnings about the missing extension and unprotected passwords on Linux, and probe diagnostics that check for a version mismatch.
FAQ
Can the server change anything in the base?
No, a change request is rejected before it reaches the base; this is a product property, not just a permissions limit.
What if the agent invents the query text itself?
The query tool only runs text that query_build assembled in the same session; arbitrary model-written text is rejected.
Related
166 skills for scientific work: bioinformatics, cheminformatics, clinical data, geospatial analysis and 100+ databases
Google's open-source MCP server for databases: ready tools for Postgres, MySQL, BigQuery, Spanner and more, plus custom tools in tools.yaml
Official Hugging Face skills: Hub operations via the hf CLI, datasets, model training, Spaces, evals and deployment
A visualization language for agents and an MCP server: neat charts from a simple semantic spec