Yandex 360 MCP Server

MCP server that gives AI assistants (Claude Code, Claude Desktop, Cursor, VS Code and any other MCP client) tools to work with Yandex 360 organization services — starting with Yandex Wiki. Transport is stdio, the only runtime dependency is mcp, and it installs & runs through uvx without cloning.

🇷🇺 Документация на русском — README_ru.md.

Scope: this server covers Yandex 360 services (Wiki, and next Directory, Disk). Yandex Tracker is intentionally out of scope — it already has an excellent, mature server: aikts/yandex-tracker-mcp. Use the two side by side.

Features

• 📖 Yandex Wiki — read and edit pages (YFM), full-text search, page tree navigation, create / update / append / delete.

• 🧩 Modular by service — each Yandex 360 service is a separate module. A module registers its tools only if its tokens are set, so the model is never handed tools for services you don't use — the context stays lean.

• 🔌 Zero-friction install — one dependency, stdio transport, runs via uvx straight from GitHub or PyPI.

• 🔐 No secrets in code or config — all tokens come from environment variables.

Modules

Related MCP server: lark-mcp

MCP Client Configuration

Prerequisites

Install uv (provides uvx):

curl -LsSf https://astral.sh/uv/install.sh | sh

Get a Yandex OAuth token with wiki:read / wiki:write scopes and your Yandex 360 organization ID. Then add the server to your client using one of the blocks below.

Add to your project's .mcp.json (or run claude mcp add):

{
  "mcpServers": {
    "yandex360": {
      "type": "stdio",
      "command": "uvx",
      "args": ["yandex360-mcp@latest"],
      "env": {
        "WIKI_TOKEN": "${WIKI_TOKEN}",
        "WIKI_ORG_ID": "${WIKI_ORG_ID}"
      }
    }
  }
}

To track the unreleased main branch instead, use args: ["--from", "git+https://github.com/pa1ch/yandex360-mcp@main", "yandex360-mcp"].

One-click install: download the latest yandex360-mcp-*.mcpb from Releases, open it in Claude Desktop, and fill in the token fields — no JSON editing. Requires uv installed on the machine.

Or configure manually — edit claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/, Windows: %APPDATA%\Claude\):

{
  "mcpServers": {
    "yandex360": {
      "command": "uvx",
      "args": ["yandex360-mcp@latest"],
      "env": {
        "WIKI_TOKEN": "your-oauth-token",
        "WIKI_ORG_ID": "your-org-id"
      }
    }
  }
}

Restart Claude Desktop after saving.

Add to .cursor/mcp.json (project) or ~/.cursor/mcp.json (global):

{
  "mcpServers": {
    "yandex360": {
      "command": "uvx",
      "args": ["yandex360-mcp@latest"],
      "env": {
        "WIKI_TOKEN": "your-oauth-token",
        "WIKI_ORG_ID": "your-org-id"
      }
    }
  }
}

Add to .vscode/mcp.json:

{
  "inputs": [
    { "id": "wiki_token", "type": "promptString", "description": "Yandex Wiki OAuth token", "password": true },
    { "id": "wiki_org_id", "type": "promptString", "description": "Yandex 360 organization ID" }
  ],
  "servers": {
    "yandex360": {
      "type": "stdio",
      "command": "uvx",
      "args": ["yandex360-mcp@latest"],
      "env": {
        "WIKI_TOKEN": "${input:wiki_token}",
        "WIKI_ORG_ID": "${input:wiki_org_id}"
      }
    }
  }
}

Any stdio MCP client works. Point it at the command:

uvx yandex360-mcp@latest

with WIKI_TOKEN and WIKI_ORG_ID in the environment.

Available MCP Tools

Configuration

All configuration is via environment variables — nothing sensitive lives in code or config files.

* A module activates only when all of its variables are set. Unconfigured modules are skipped (the server logs this to stderr and keeps running).

See .env.example for a template.

Loading only the modules you need

Every registered tool takes up space in the model's context, so this server exposes only the modules you actually configure — nothing else reaches the model. At startup a module's tools are registered only when both conditions hold:

1. all of the module's environment variables are set (e.g. wiki needs WIKI_TOKEN and WIKI_ORG_ID), and

2. the module is not excluded by YANDEX360_ENABLE.

Two ways to control the exposed tool surface:

• By configuration (default). Set tokens only for the services you want. Any module whose variables are missing is silently skipped. Configure just wiki → only the nine wiki_* tools are exposed, and nothing for directory / disk.

• By explicit allow-list. Set YANDEX360_ENABLE to a comma-separated list to pin an exact subset even when more modules are configured — e.g. YANDEX360_ENABLE=wiki.

The server logs which modules are active (and which were skipped) to stderr on startup.

For AI agents installing this server: keep the tool surface minimal — configure only the services the user actually needs. Provide just those modules' tokens (and/or pin YANDEX360_ENABLE). Do not add tokens for services that won't be used: every extra module loads tools that only waste the model's context.

Development

git clone https://github.com/pa1ch/yandex360-mcp.git
cd yandex360-mcp
uv sync                       # create venv + install deps
uv run yandex360-mcp          # run the server over stdio
uv run pytest                 # run tests

Publishing to PyPI

uv build                      # wheel + sdist into dist/
uv publish                    # needs a PyPI token (UV_PUBLISH_TOKEN or --token)

Roadmap

• directory module — users, departments, org structure (Yandex 360 Directory API).

• disk module — file and folder operations (Yandex Disk API).

Contributions welcome — a new service is a single module exposing is_configured() and register(mcp), then one line in the module registry.

License

MIT