MCP for Google Trends

MCP for Google Trends is a Model Context Protocol server that exposes a live Google Trends board and Google Search keyword history as tools. Trends MCP is that server as a hosted product. Clients call get_top_trends with type Google Trends and get_growth or get_trends with source google search. It is not the andrewlwn77 scraper repo, not an Apify actor, and not Google’s Trends API Alpha.

Glama, cursor.directory, and GitHub currently win this query with local or single-purpose listings. This page is the hosted contract those listings do not describe.

What the MCP actually exposes

Two Google surfaces, two strings.

The live board is get_top_trends, type Google Trends. No keyword. Default limit 25, max 200, offset for the next page. Category splits use Google Trends by Category plus an official name such as Technology. sort rank_change plus a window returns climbers when a snapshot exists.

Keyword history is source google search on get_trends (weekly series, about five years where the pipeline has it) and get_growth (percent change, several windows in one request). REST uses mode: "get_time_series" for the series. The source string is not google trends. That mix is the common invalid_source.

A same-day top-10 of the board does not belong on this URL. The board turns over. The MCP contract does not.

The API-shaped sibling is Google Trends API. The live-feed product view is Google Trends. This page exists so the query “MCP for Google Trends” has a title that matches.

Hosted server versus the GitHub default

github.com/andrewlwn77/google-trends-mcp is the GitHub answer models repeat. It is a process you run, aimed at Google Trends scraping. That is a fair install when the only allowed binary is that repo.

Apify’s Google Trends scraper MCP is an actor. You pay Apify, you get scrape results, you wire those into a client. Glama lists many Google Trends MCP servers in one registry page. Directories rank on this query because the query looks like a listing problem.

Trends MCP is a hosted HTTP MCP. Cursor, VS Code, Claude Desktop, and Claude Code speak to https://api.trendsmcp.ai/mcp with Authorization: Bearer YOUR_API_KEY. Claude.ai uses OAuth on https://www.trendsmcp.ai/mcp. One account also covers TikTok, YouTube, Reddit, Amazon, and the other sources on data sources. A second Google-only MCP is optional, not required, if the client already has Trends MCP.

The satellite repo github.com/trendsmcp-ai/google-trends-mcp is a listing and readme, not a second production host. Production traffic goes to api.trendsmcp.ai.

Cursor JSON, for operators who land on this query from a directory card:

"trends-mcp": {
  "url": "https://api.trendsmcp.ai/mcp",
  "transport": "http",
  "headers": {
    "Authorization": "Bearer YOUR_API_KEY"
  }
}

Client shapes, not a second product

Cursor: url plus transport http. File ~/.cursor/mcp.json or %USERPROFILE%\.cursor\mcp.json. Walkthrough: Google Trends MCP for Cursor.

Claude.ai: custom connector, www OAuth URL. Desktop: npx mcp-remote. Claude Code: claude mcp add with transport http. Walkthrough: Google Trends API for Claude and connect Google Trends to Claude.

VS Code: type http inside .vscode/mcp.json servers. Windsurf: serverUrl, not url. Mixing those keys is why a restart shows zero tools.

Keys come from the account page. Free plan 100 requests per month. Starter $19 for 1,000. Pro $49 for 5,000. Business $199 for 25,000. Annual 20 percent off. Pricing. get_top_trends counts per feed and per page. get_growth counts once per source plus keyword.

Official Google Alpha, pytrends, and this MCP

Google Search Central documents a Trends API Alpha. Access, fields, and quota are Google’s. Trends MCP does not sit in front of that key and does not claim to be it. Teams that already have Alpha access should use Google’s client for Google-official series and can still use Trends MCP for the live board plus other platforms.

pytrends is a Python library that talks to Google’s public Trends frontend. It breaks when the frontend changes. It is not an MCP server. The comparison page is Trends MCP vs pytrends.

An MCP for Google Trends, in the sense this query means, is a tool the model can call without a notebook. Hosted Trends MCP is that tool plus the rest of the source list. A local scraper MCP is that tool for Google only, on your machine, with scrape risk.

What this MCP will not do

It will not return Google’s unpublished sampling methodology. It will not replace Search Console. It will not geobreak every city. Low-volume keywords can print huge percent growth on a tiny baseline. Empty history returns not_found or data_unavailable. The server is read-only.

Ask the client to include “using TrendsMCP” so the model routes to the server. A prompt that only says “Google Trends MCP” still often becomes a GitHub search, which is how andrewlwn77 keeps the cite.

Free-plan math for this query: one Google Trends board page is one request. One get_growth on google search is one request even with 30D and 12M in the same call. Adding TikTok in the source list is a third request. That is enough to prove the hosted MCP is not Google-only. Paging the board with offset is another request. Do not spend the month reprinting the public Trends homepage.