Content
# mcp-1c
[](https://safeskill.dev/scan/feenlace-mcp-1c)
MCP Server for integrating AI assistants with 1C:Enterprise.
AI sees your 1C configuration metadata and generates accurate code in BSL. Works with any MCP-compatible client.
### Works with Local Models
MCP-1C is not tied to a specific neural network. Works with any MCP-compatible client:
- **Local Models** (Ollama, LM Studio, llama.cpp) - data does not leave your network
- **Cloud Services** (Claude, GPT, YandexGPT, GigaChat) - through corresponding MCP clients
- **IDEs with AI** (Cursor, VS Code + Continue/Cline, JetBrains)
Your code and 1C data remain with you. MCP-1C is a local process that communicates only with your database.
## Paid Versions
In addition to the free Open version, paid editions with extended capabilities are available:
- **Extended** (1,990 ₽/month) — 7 tools: query optimizer, BSL linter, syntax assistant, multi-base support, .cfe extensions, and more
- **Professional** (4,990 ₽/month) — all from Extended plus: semantic search, dependency graph, mass code analysis, security audit, code base metrics, and update assistant
When registering, you receive 14 days of the Professional version for free.
→ [More about tariffs](https://feenlace.ru/pricing)
→ [Documentation](https://feenlace.ru/docs)
> [!TIP]
> **Beta test of the Professional version.** We have launched the Professional edition with 6 tools for in-depth code base analysis: semantic search, dependency graph, mass analysis, security audit, quality metrics, and update assistant.
>
> Register and get 14 days free. Active beta testers who provide useful feedback and want to continue testing will have their trial period extended. Top-5 beta testers will receive a free lifetime subscription to the Professional version.
>
> → [Register](https://feenlace.ru/register) | [Pro Documentation](https://feenlace.ru/docs#pro-tools) | [Report an issue](https://github.com/feenlace/mcp-1c/issues)
## Version Comparison
| | Open | Extended | Professional |
|---|---|---|---|
| Tools | 9 | 7 | 7 + bulk_analyze |
| Price | Free | 1,990 ₽/month | 4,990 ₽/month |
| Trial Period | — | — | 14 days |
| License | MIT | Subscription | Subscription |
**Extended adds:**
- Query optimizer (15 antipatterns)
- BSL linter (30+ diagnostics)
- Syntax assistant (10,000+ definitions)
- Multi-base support and .cfe extensions
- Code sandbox, query generator, and print form generator
- Modal call converter to asynchronous
- Smart context and data auto-update
**Professional adds:**
- Semantic search in code (LSA + Randomized SVD)
- Module dependency graph (10 types of connections, SQLite)
- Mass code analysis (antipatterns, duplicates, dead code)
- BSL security audit (11 SEC rules)
- Code base metrics (LOC, complexity, technical debt 0-100)
- Update assistant (3-way diff)
- CI/CD integration (--ci, --json, quality gates)
- HTML/PDF reports
→ [Register](https://feenlace.ru/register) | [Tariffs](https://feenlace.ru/pricing)
## Why mcp-1c
- **One binary, zero dependencies.** Written in Go — no need for Python, Node.js, JVM, or EDT. Downloaded, launched, and works.
- **9 tools for working with a live database.** Metadata, configuration information, forms, data queries (with parameters), code search, validation, registration log, and BSL reference.
- **Full-text code search** (`search_code`). Three modes: smart (BM25 ranking), regex, exact. Built-in BSL synonyms — search for `StrFind` finds `СтрНайти` and vice versa.
- **Shard indexing.** Parallel index construction by the number of cores. ~7 seconds for 13,000+ modules. Disk cache — repeated launch is instant.
- **Non-blocking start.** Index is built in the background, MCP server is available immediately. Search will work after indexing is complete.
- **Works with your database.** AI sees the real configuration and real data — not abstract documentation, but your actual database.
- **Not tied to IDE and neural network.** Works with Configurator, EDT, or without IDE. Works with any model, including local ones (Ollama, LM Studio). Only HTTP 1C service is required.
- **Auto-installation.** `mcp-1c --install "C:\path\to\database"` — will find the platform, install the extension, and update the database configuration.
- **Built-in BSL reference.** Syntax of platform functions is available without external services and without launched 1C.
## Quick Start
> **Hearing about MCP for the first time?** Read the [step-by-step guide](docs/getting-started.md) — it covers everything from scratch, including what MCP is.
### 1. Download
Binary for your OS — in [Releases](../../releases). Or: `go build -o mcp-1c ./cmd/mcp-1c/`
### 2. Install Extension in 1C
```bash
# Windows
mcp-1c --install "C:\path\to\database"
# macOS / Linux
mcp-1c --install ~/Documents/InfoBase
# Client-server database (MS SQL, PostgreSQL)
mcp-1c --install "srv-1c\buh_prod" --server --db-user Admin --db-password pass
```
> If the platform is installed in a non-standard folder: `mcp-1c --install "path" --platform "/custom/path/to/1cv8"`
>
> If the platform version is not automatically determined (non-standard path without version number), specify it explicitly: `mcp-1c --install "path" --platform "/custom/path/to/1cv8" --platform-version 8.3.13`
### 3. Launch 1C HTTP Service
**Recommended method** — standard publication via Apache or IIS (Configurator → Administration → Publication on web server). Works on Windows and Linux. Details — in the [step-by-step guide](docs/getting-started.md#step-3-launch-1c-http-service).
**Quick launch for development (only Windows):**
```cmd
"C:\Program Files\1cv8\8.3.XX.XXXX\bin\1cv8.exe" ENTERPRISE /F "C:\path\to\database" /HTTPPort 8080
```
> The `/HTTPPort` parameter is undocumented, works **only on Windows** and is suitable for quick verification. On Linux, use Apache or ibsrv, on macOS, 1C HTTP services are not supported.
### 4. Configure AI Client
MCP server configuration is the same for any client and any model. It doesn't matter if you use Claude, Ollama, or LM Studio; MCP-1C configuration remains the same:
```json
{
"mcpServers": {
"1c": {
"command": "/path/to/mcp-1c",
"args": ["--base", "http://localhost:8080/hs/mcp-1c"]
}
}
}
```
> On Windows, paths with backslashes: `"command": "C:\\path\\to\\mcp-1c.exe"`
Restart the AI client. In Claude Desktop, we recommend: **«+» → Connectors → Tool access → Always available**.
> Also supported: **Claude Code**, **Cursor**, **Windsurf**, **VS Code + Copilot**, **VS Code + Continue**, **JetBrains IDE**, as well as any clients for local models with MCP support. Configuration of each — in the [step-by-step guide](docs/getting-started.md#step-4-configure-ai-client).
Ask: **«Show the structure of my 1C database configuration»**
## Available Tools
| Tool | Description |
|------------|---------|
| `get_metadata_tree` | Metadata tree: catalogs, documents, registers, common modules, etc. |
| `get_object_structure` | Attributes, table parts, measurements, and resources of a specific object |
| `get_form_structure` | Structure of the main form: elements, commands, event handlers |
| `get_configuration_info` | Configuration name, version, supplier, platform version, operating mode |
| `search_code` | Full-text search in module code: smart (BM25), regex, exact. BSL synonyms (Russian↔English). Filtering by metadata type and module |
| `bsl_syntax_help` | Reference for 180 built-in functions, type methods, and BSL patterns |
| `execute_query` | Execute a 1C query with parameters (only SELECT) |
| `validate_query` | Check query syntax without execution |
| `get_event_log` | Reading the registration log with filtering by date, level, and user |
## Configuration
| Flag | Env var | Default | Description |
|------|---------|-------------|----------|
| `--base` | `MCP_1C_BASE_URL` | `http://localhost:8080/hs/mcp-1c` | URL of the 1C HTTP service |
| `--user` | `MCP_1C_USER` | — | User of the HTTP service |
| `--password` | `MCP_1C_PASSWORD` | — | Password of the HTTP service |
| `--dump` | — | — | Path to configuration dump (DumpConfigToFiles), includes search_code tool |
| `--reindex` | — | — | Forced reindexing of the search index (ignores cache) |
| `--install` | — | — | Install extension to the 1C database at the specified path |
| `--server` | — | — | Client-server mode: `--install` accepts connection string `server\database` (e.g., `srv-1c\buh_prod`) |
| `--platform` | — | — | Path to the 1C binary (auto-detected if not specified) |
| `--platform-version` | — | — | 1C platform version (e.g., `8.3.13`). Automatically detected from the platform path. Specify manually if the platform is installed in a non-standard path without version information. Minimum supported version: 8.3.10 |
| `--db-user` | — | — | 1C database user for DESIGNER ( `--install` mode) |
| `--db-password` | — | — | 1C database password for DESIGNER ( `--install` mode) |
## Logging and Output
By default, behavior depends on whether the server is run in a terminal or through an MCP client:
- **In terminal** (stdin connected to tty): indexing progress, informational messages, and errors are written to stderr as usual.
- **Through MCP client** (Kilo Code, OpenCode, Claude Desktop, Cursor, etc., when stdin is a pipe): stderr is empty, random output from third-party libraries is redirected to `~/.cache/mcp-1c/stderr.log`. This protects clients that interpret any stderr output as a fatal error (Issue #14).
### Flags and Environment Variables
| Flag / env | Description |
|------------|---------|
| `--verbose` | Force enable stderr even when running through a pipe. Convenient for debugging MCP client connections. |
| `--quiet` | Force disable stderr even in terminal. Overrides `--verbose`. |
| `MCP_1C_NO_TTY=1` | Equivalent to `--quiet`. More convenient for CLI flag when running in Docker / systemd where command-line arguments are less flexible. |
| `--debug` | Detailed logs to file `~/.cache/mcp-1c/server.log`. Also disables progress indicator in terminal. |
### Git Bash / MSYS2 / MinTTY on Windows
These shells connect stdin through named pipes, not through a regular console handle. Autodetect considers them non-TTY, so indexing progress is not output by default. For manual diagnostics, use the `--verbose` flag or a full `cmd.exe` / Windows Terminal.
## Development
```bash
go build -o mcp-1c ./cmd/mcp-1c # build
go test ./... -v -race # tests
go run ./cmd/mock-1c -port 9191 # mock 1C server
```
### 1C Extension
Extension sources are stored in `extension/src/` in XML configuration dump format. With `--install`, they are embedded into the binary through `go:embed` and loaded directly through DESIGNER `/LoadConfigFromFiles`. No ready `.cfe` file is required for building.
For manual installation without CLI, build `.cfe` from sources:
```bash
# macOS / Linux (requires installed 1C platform)
./scripts/build-extension.sh ~/Documents/InfoBase
# Windows
scripts\build-extension.cmd C:\Users\User\Documents\InfoBase
```
## Compatibility
| AI clients | |
|------------|---|
| **Local models** | Ollama, LM Studio, llama.cpp, and any MCP-compatible clients |
| **Cloud services** | Claude Desktop, Claude Code, GPT (via MCP client), YandexGPT, GigaChat |
| **IDE** | Cursor, VS Code (Continue, Cline, Copilot), Windsurf, JetBrains IDEs |
MCP-1C does not know or determine which model is working on the client-side. The configuration is the same.
| 1C Platform | Status |
|-------------|--------|
| 8.3.10 and above (commercial) | Supported |
| 8.5.x (commercial) | Supported |
| 8.3.10+ / 8.5.x (educational) | Supported |
> Minimum supported platform version: **8.3.10**
| OS | MCP Server | Auto-install | 1C HTTP Service |
|----|-----------|---------------|----------------|
| Windows | yes | yes | yes (Apache/IIS recommended, /HTTPPort for quick start) |
| macOS | yes | yes | no (1C platform limitation), use Windows-VM |
| Linux | yes | yes | yes (Apache or ibsrv; /HTTPPort not available) |
## Publications
[](https://infostart.ru/1c/articles/2635907/)
- [One binary, zero dependencies: connecting AI to your 1C database in 10 minutes](https://infostart.ru/1c/articles/2635907/)
## License
MIT
