Yandex Webmaster MCP
This MCP server lets you query Yandex Webmaster data in plain language, covering search queries, indexing, diagnostics, sitemaps, backlinks, and recrawl status. It is read-only by default (except recrawl submission) and uses OAuth without storing secrets.
README
Yandex Webmaster MCP
Ask your Yandex Webmaster data in plain language — from Claude, Cursor, or any MCP client.
Read-only by design (plus URL recrawl), no secrets stored. Sign in with a built-in public OAuth client over PKCE — no app registration, no client secret.
Early development (v0.1). The tool surface is still growing.
A Model Context Protocol server that exposes Yandex Webmaster as a small set of
flexible, well-described tools instead of one-thin-wrapper-per-endpoint. Pairs
with yandex-metrica-mcp:
run both and an agent can cross-reference search queries from Webmaster with
on-site behavior and conversions from Metrica in a single conversation.
Quickstart
1. Add the server to your MCP client (Claude Desktop, Cursor, …):
{
"mcpServers": {
"yandex-webmaster": {
"command": "npx",
"args": ["-y", "@boxlab/yandex-webmaster-mcp"]
}
}
}
2. Sign in — interactive, no secret needed:
npx @boxlab/yandex-webmaster-mcp auth
Approve access in the browser, paste the code Yandex shows you. The token is
cached at ~/.config/yandex-webmaster-mcp/token.json (mode 0600).
3. Ask your agent — e.g. "List my Webmaster hosts and show the SQI for example.com."
Or install as a Claude Code plugin
/plugin marketplace add BoxLab-Ltd/yandex-webmaster-mcp
/plugin install yandex-webmaster-mcp@boxlab
Why
Existing community servers tend to be thin wrappers — one tool per API endpoint, dumping raw JSON into the model's context. This server aims to be well-engineered: fewer, smarter tools; compact, structured output; read-only by default so an agent can explore your Webmaster data safely.
Tools (v0.1)
get_hosts— list the sites available to your token with theirhost_id, URL and verification state; optionally a host summary (SQI, indexed/excluded pages, site problems). Read-only. Call this first.search_queries— search-query analytics from Yandex Search: impressions, clicks, average show/click position.report="top"ranks the queries bringing the most traffic;report="trend"returns a time series (per query or the site aggregate). Filter by device and date range. Read-only.get_indexing— how Yandex crawls and indexes the host.report="history"gives crawled pages by HTTP status class over time;report="crawled"lists example crawled URLs with their codes;report="in_search"lists example pages currently in search. Read-only.get_diagnostics— problems detected on the site (DNS, slow response, robots.txt, 4xx, …) with severity, active problems first. Read-only.list_sitemaps— the Sitemap files Yandex knows, with URL counts, error counts and where each was discovered. Read-only.get_external_links— inbound links (backlinks):report="samples"for example links with source/target,report="history"for the count over time. Read-only.recrawl_status— remaining daily recrawl quota plus recent recrawl tasks and their state; passtaskIdto check one task. Read-only.recrawl_submit— ask Yandex to recrawl a specific URL sooner. The one write tool; consumes one unit of the daily quota. Returns the task id.
Planned: write tools for sitemaps, original texts and feeds, deferred to a later, opt-in release.
Requirements
- Node.js >= 18
- A Yandex account with at least one site added in Yandex Webmaster
Authentication
Three ways, in priority order:
- Interactive login (recommended) —
npx @boxlab/yandex-webmaster-mcp auth. Uses the built-in public OAuth client over PKCE; the token is valid ~6 months. - Static token — set
YANDEX_WEBMASTER_TOKEN(e.g. for CI) to an OAuth token for an app with thewebmaster:hostinfoscope. - Your own OAuth app — set
YANDEX_OAUTH_CLIENT_ID(andYANDEX_OAUTH_CLIENT_SECRETto enable automatic token refresh). The app needs thewebmaster:hostinfoandwebmaster:verifyscopes.
See .env.example for all configuration.
Development
Bun-first:
bun install
bun run dev # watch mode
bun run typecheck
bun run lint
bun test
bun run build # emits dist/ (Node-compatible)
bun run try # smoke against the real API after `bun run auth`
The published package runs on Node; local development uses Bun.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。