1C AI Workbench
A local read-only MCP server suite for 1C: a Rust code indexer, metadata skills, a prompt gallery and platform help search, with evidence-backed answers
Low risk
We rate an entry low when it mostly gives the agent instructions and reference material.
Why this level
- The default mode is read-only, and the export is mirrored rather than modified by indexing
- The only write path, 1c-ibcmd, is disabled by default and needs two separate explicit confirmations
Install
Manual install
.\scripts\setup.ps1
.\scripts\04_index_1c_dump.ps1 -DumpRoot "C:\1c-ai-client\dump" -Force
.\scripts\06_healthcheck.ps1Installs dependencies, indexes the export and checks readiness after downloading the signed release.
This is third-party code. Review the repository files before installing.
What it does
The workbench mirrors a 1C configuration's XML export into a read-only copy, indexes BSL code and metadata with the Rust bsl-indexer binary, and serves the result through several MCP servers: 1c-code-index (the main SQLite-backed code index), 1c-skills (16 read-only tools like meta_info, form_info, skd_info, role_info, cf_info), 1c-prompt-gallery (the explain-module, find-object, review-bsl, trace-flow prompts as callable tools) and 1c-help-index (local search over the platform's .hbk help via FTS5). An experimental 1c-ibcmd bridge for live ibcmd operations is disabled by default and requires explicit IBCMD_ALLOW_WRITE=1 and confirm_replace=true. The stated answer contract requires the AI to give a short conclusion, the 1C object's name and type, the file path, the module or procedure, an evidence snippet, a confidence level, and manual verification steps in the Configurator or EDT. Production builds go through a two-phase process: a signed build, verification of the same files on a clean Windows machine, then publication without a rebuild.
Who it is for. For 1C teams who need fast, grounded answers about a configuration (where a register is written, which object a form belongs to, what changed between exports) with evidence instead of guessing from code fragments.
Good fit when
- You need a fast, grounded answer about the structure of a large 1C configuration
- You need to search the platform help locally, with no network access
- You want to give an agent read-only access to code and metadata without risking accidental writes to the database
Not a fit when
- You need to run on something other than Windows: the workbench is built for Windows 10/11 with PowerShell
- You need live database operations via ibcmd: they are experimental, disabled by default, and require an explicit write permission
Example request
Find where the inventory balance register is written and show the module path with a code snippetLimitations
Only runs on Windows 10/11 with PowerShell 5.1+, Git for Windows and Python 3.10+ for the Python bridges. Needs a signed installer with a prebuilt bsl-indexer.exe or a Rust toolchain to build from source. The experimental ibcmd bridge for database operations is deliberately disabled by default. The project has a commercial side: premium prompts, golden answers and support are available separately from the maintainer.
How to disable. Stop the running workbench MCP servers and remove their entries from the MCP client config (configs/*.json); the workbench itself can just be deleted from disk.
MCP
- Transport
- stdio
- Authentication
- API key
| Environment variables | |
|---|---|
| IBCMD_ALLOW_WRITE | Explicitly enables experimental write operations via ibcmd; disabled by default. |
Security check
- The default mode is read-only, and the export is mirrored rather than modified by indexing
- The only write path, 1c-ibcmd, is disabled by default and needs two separate explicit confirmations
README in short
The README gives a readiness status with test metrics (67/67 pytest, 8 PASS in the e2e smoke test, 6/6 healthcheck), explains the architecture from the XML export through the mirror and Rust indexer to four Python bridges, requirements, and a quick start via a set of numbered scripts. A section per MCP server gives its launch command, a response-contract section lists the required fields of a good AI answer, and a security model section is described separately (read-only by default, API keys owned by the client, no bundled proprietary binaries). It names its borrowing sources (cc-1c-skills, BSL Language Server, OneScript, Vanessa Runner) and the maintainer's commercial offering. MIT license, REUSE-compliant repo.
FAQ
Can the workbench change the database or the export?
No by default: the default mode is local and read-only, the export is mirrored before indexing, and the live write path via ibcmd is disabled and needs explicit permission.
Does it bundle proprietary 1C binaries?
No, the README explicitly states no proprietary 1C binaries are bundled with the workbench.
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