Yandex Cloud docs in MCP: offline SQLite search
yandex-cloud-docs-mcp
An offline MCP server with full-text search over a local copy of the Yandex Cloud docs: no network, read-only, line-numbered citations
- 0 ★ on the repository
- Last commit
Minimal access
Why this level
- The server only reads local documentation files, with no network and no authorization
Install
- Works in
- Services
Manual install
claude mcp add --transport stdio --scope user yandex-cloud-documentation -- "$NODE_BIN" "$MCP_SERVER" --docs-root "$DOCS_ROOT" --index "$DOCS_INDEX"First clone `yandex-cloud/docs` and this repository, run `npm ci`, `npm run build` and `npm run index`; the paths in the variables are absolute.
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The server only reads local documentation files, with no network and no authorization
This is third-party code. Review the repository files before installing.
What it does
A TypeScript MCP server builds a SQLite FTS5 index over a local clone of the `yandex-cloud/docs` repository (the `md-docs` folder) and answers the agent with three tools: `search_yandex_cloud_docs` searches and returns paths, excerpts and line numbers, `read_yandex_cloud_doc` reads up to 400 lines of a document, and `yandex_cloud_docs_status` compares the index revision with the current docs revision. The server does not touch the network and needs no authorization, so it fits closed environments. The index updates from Git history for changed files only, and a full rebuild is atomic. Protection against leaving the folder and against symlinks is built in. It connects to Claude Code, Claude Desktop and Codex over stdio.
Who it is for. Engineers and devops on Yandex Cloud who need an agent to answer from the docs with exact citations and without internet access.
Good fit when
- The agent has to find the right Yandex Cloud docs section and cite it with lines
- Work happens in a closed environment and reaching out for docs is forbidden
- You need fast search across thousands of pages without hitting the web version
- The docs keep changing and the index should catch up through Git history
Not a fit when
- You need to manage cloud resources: the server only reads docs and calls no cloud API
- You do not want to clone the docs repository and build an index: the server is empty without it
- You need search over other vendors docs: the index is built for Yandex Cloud only
Example request
Find in the Yandex Cloud docs how to create an Object Storage bucket with versioning and show a quote with linesMCP
- Transport
- stdio
- Authentication
- not required
Limitations
Before use you must clone `yandex-cloud/docs`, run `npm ci`, `npm run build` and build the index with `npm run index`; the first build is full and takes time. All paths in the client settings are absolute. The corpus is mostly Russian, and the server asks the client to translate terms for search. There is no ready `.mcpb` extension for Claude Desktop, and it is installed by hand as a stdio server. The project is small, zero stars, last commit in September 2026, Apache-2.0 license.
How to disable. Remove the server with `claude mcp remove yandex-cloud-documentation` or the entry in the client config, and the local clone and index can be erased.
FAQ
Is the internet needed to work?
No: the server reads a local index and a local docs copy, with no network requests or authorization.
Where do the docs come from?
From a clone of the `yandex-cloud/docs` repository; the index is built over its `md-docs` folder.
How do I update the index after the docs update?
Run `git pull` in the docs clone and repeat `npm run index`: changed Markdown files are picked up from Git history without a full rebuild.
How many tools does the server have?
Three: search, reading a document and the index status.
Can it read outside the docs folder?
No, the server is protected against path traversal and symlinks that escape the root.
Discussion
No questions or reviews yet.
Related
All Yandex Cloud toolsYandex Cloud's official MCP server set: documentation, search, app deployment, serverless functions, containers and a data catalog
Read-only Yandex Cloud MCP server
Yandex Cloud MCP Server
An unofficial read-only Yandex Cloud MCP server: VMs, networks, disks, and access at folder, cloud and org level
Yandex search MCP server in TypeScript
yandex-search-mcp
A TypeScript MCP server for the Yandex Search API with its own result parser and tests
Yandex Cloud MCP: VMs and Object Storage
yandex-cloud-mcp
MCP server for basic Yandex Cloud operations: virtual machines, Object Storage buckets and serverless functions from an agent chat
Yandex Cloud Object Storage MCP server
yandex-cloude-mcp
An unofficial single-file MCP server for Yandex Object Storage: bucket listing, contents and file downloads via the AWS SDK
An MCP server for the Yandex Cloud Search API v2: Wordstat and web search with file caching, date normalization and 429-retry handling
README in short
The README sets out features, quick installation from two repositories, absolute paths, connection to Claude Code, Claude Desktop and Codex, a smoke test, index updates and the security model.