Gyrfalcon: 1C code navigation MCP
gyrfalcon
A single dependency-free Rust binary with no network calls: indexes a 1C configuration export and answers call-graph and metadata questions in milliseconds
- 16 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The CLI parses a 1C configuration export and writes an index to a local SQLite database
- The install command writes MCP, skill and role settings into the agent config; --plan previews the changes
Install
- Works in
- Services
Manual install
gyrfalcon build <путь к src> --out config.dbBuilds the index from a configuration export; the binary comes from GitHub releases.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The CLI parses a 1C configuration export and writes an index to a local SQLite database
- The install command writes MCP, skill and role settings into the agent config; --plan previews the changes
This is third-party code. Review the repository files before installing.
What it does
The server builds a deterministic index of a 1C configuration by parsing BSL with tree-sitter instead of having the agent re-read the export. It builds a call graph from the AST with a confidence class on every edge, and 27 metadata tables: attributes with types, register movements, subscriptions, role rights, and extension overrides with both sides' addresses. Twelve MCP tools answer structural questions, plus a coverage tool that honestly reports what the index does not know. The install command detects an installed harness (Claude Code, Kilo, Codex) itself and writes the MCP config, a skill and roles, and --plan shows what would change without changing anything.
Who it is for. For 1C developers on large configurations who need fast, precise answers about code structure without the agent re-reading the export.
Good fit when
- You work with a large configuration export and the agent is slow reading files directly
- You need a call graph or a precise breakdown of extension overrides
- You need everything to run fully locally, with no network and no API key
Not a fit when
- The configuration is small and the agent can just read files directly
- You need harness support beyond Claude Code, Kilo and Codex
Example request
Use gyrfalcon to find every place this procedure is overridden in extensions and show the call graphMCP
- Transport
- stdio, http
- Authentication
- not required
Limitations
It needs a file export of the configuration; the server does not analyze a live infobase directly. The index for the heaviest tested configuration is 3.8 GB and takes 2 to 2.5 minutes to build, worth accounting for on very large bases. The install command edits the agent's config, so it is worth checking the plan first with --plan.
How to disable. Remove the gyrfalcon entry from the agent's MCP config and the skill folder, for example ~/.claude/skills/gyrfalcon/ for Claude Code.
FAQ
Do I need Rust or Python to use it?
No, not if you download the prebuilt binary from releases. Rust is needed only to build from source.
Does the server send code anywhere?
No, there are no network calls either at index build time or when answering queries; everything stays on the machine.
Discussion
No questions or reviews yet.
Related
All 1C:Enterprise tools1C configuration graph MCP server
1C Configuration MCP Server
An MCP server builds a metadata and call graph from a 1C configuration's XML and BSL export, with optional semantic search via an external embedder and Qdrant
1C AI development environment
1C AI Development Environment
1C development environment with an MCP server offering 10 tools: 11 BSL analyzers, JSON DSL to XML, dependency graph, and EPF build without 1C
A single-binary Go MCP server: the agent sees 1C metadata, searches code, runs read-only queries and writes accurate BSL
An MCP server for Claude Code: full-text search over multiple 1C XML+BSL configuration exports in a local SQLite FTS5 database
A native component that runs a full MCP server inside 1C: tools, resources and prompts are written in BSL, while a DLL handles the protocol
Five independent Docker-based MCP servers: a Neo4j metadata and call graph, static BSL analysis, platform syntax help, a query builder and 1C coding standards
README in short
The README gives benchmarks on the heaviest tested configuration: 98,117 files, 752,335 methods, a full index build in 132 to 160 seconds with a 20 percent spread. It covers the quick start with a prebuilt binary, building from source with cargo, twelve MCP tools, and installing into Claude Code, Kilo and Codex through a single install command with a --plan preview flag.