Purpose
mcp-yandex-wordstat is an MCP (Model Context Protocol) server for connecting Yandex Wordstat search demand statistics to AI applications, enabling AI clients (such as Claude, Cursor, Codex, and other AI clients) to directly query and interpret keyword search demand.
You can ask how often a phrase is searched, which months show rising interest, and which cities favor a topic more, and the assistant will collect Wordstat data and interpret the results.
Main capabilities
- Provides five tools: top and similar queries, demand dynamics, regional distribution, region directory, and a technical raw API request tool.
- Queries top queries and semantically similar queries; totalCount corresponds to the last 30 days.
- Dynamics returns a {date, count, share} time series, supporting daily, weekly, or monthly granularity.
- Regions breaks down the last 30 days of demand by region; an affinityIndex above 100% indicates above-average interest.
- Supports narrowing and comparing reports by region and device (desktop, mobile, tablet).
- Read-only: the Wordstat API does not create campaigns, ads, keyword phrases, or other objects.
- Counter values may be returned as strings because Yandex transmits large integers in this form in JSON.
Usage
Requirements: Node.js 20 or newer, plus a Yandex Cloud Search API key and a Yandex Cloud catalog ID.
Adding via the AI application interface:
- Open Settings → Plugins → MCP servers.
- Click Add server.
- Add the launch command
npx -y mcp-yandex-wordstat@latestand the environment variablesWORDSTAT_API_KEY、WORDSTAT_FOLDER_ID.
Configuration for different clients (see the repository documentation for details):
- Claude Code and Codex have corresponding official/documented guidance.
- Claude Desktop: open Settings → Developer → Edit Config and add the server to
claude_desktop_config.json; if Edit Config is unavailable, on macOS edit~/Library/Application Support/Claude/claude_desktop_config.json, and on Windows edit%APPDATA%\Claude\claude_desktop_config.json. - Cursor: for all projects use
~/.cursor/mcp.json, and for the current project only use.cursor/mcp.json. - VS Code: open the command palette and run MCP: Open User Configuration, add the server to
mcp.json, and check startup with MCP: List Servers.
Obtaining access credentials:
- Create a service account in Yandex Cloud and assign it the role
search-api.webSearch.user. - Issue an API key for it with the scope
yc.search-api.execute(see the AI Studio documentation for the steps). - Find the catalog ID (folderId) on the catalog page in the Yandex Cloud console and in the page URL.
- Use the key as
WORDSTAT_API_KEYand the catalog asWORDSTAT_FOLDER_IDpassed in.
Recommended first safe query example:
- "Сколько в месяц ищут «купить велосипед» и какие есть похожие запросы?" (How many people search for "купить велосипед" per month, and what similar queries are there?)
Use cases
- Keyword selection and demand assessment: find out how many times a month a phrase is searched, and what similar queries there are; gather queries with frequencies around a topic; distinguish queries containing a specified phrase from semantically similar queries.
- Understanding seasonality: view the year-round demand for a keyword by month; observe which weeks demand for a service rises or falls; compare query dynamics on mobile and desktop.
- Regional comparison: find cities with above-average interest in a topic; compare demand in Moscow and Saint Petersburg; find the ID of a desired region and narrow the next report to that region.
Limitations and notes
- It does not use the legacy standalone Wordstat API with OAuth, but accesses Yandex Cloud Search API v2; it does not access or modify anything in Yandex Wordstat, and the five tools (including raw_request) are all read-only. Technically the API uses POST, but Wordstat has no write endpoints; the server additionally restricts arbitrary requests from being sent to other hosts.
- The API key is stored in plaintext in the MCP client configuration and should be treated like a password.
- One call builds data for only one phrase. For large lists of keywords, it is recommended to narrow the list first rather than running all queries consecutively: Yandex Cloud Search API quotas are shared for the same key.
- It retrieves aggregated search demand statistics, not data for a specific advertising account.
- Caching: in long-lived processes, the region tree is loaded only once and reused for subsequent requests.
- Retries: a single request times out after 60 seconds. After 429, 5xx, network errors, or timeouts, it retries up to three times; Retry-After is taken into account, and the delay does not exceed 30 seconds.
- No background monitoring: the server works when an AI application calls it. If the application supports scheduled tasks, periodic reports for selected phrases can be configured there.
- Anonymous telemetry: by default it sends a random installation identifier, event or tool names, and the versions of the server, Node.js, operating system, and AI client. It does not include API keys, tool parameters, user messages, demand data, or environment variable values. It can be disabled for the Ask Ads MCP server via
ASKADS_TELEMETRY=0. - No information about fees, pricing, etc. is provided in the file.