免费 Google Trends MCP:小工作室的日常趋势雷达,和收费 MCP 这样搭配最省钱

The Free Google Trends MCP: A Daily Trend Radar for Small Studios — and How to Pair It With the Paid One

Tech-Experiment #MCP#Google Trends#开源#AI Agent#内容创作#小工作室#趋势分析#免费工具
更新于
🇨🇳 中文

仓库: github.com/purahmanian/google-trends-mcp · MIT · TypeScript 安装: claude mcp add google-trends -- npx -y google-trends-mcp · 无需 API Key 数据源: Google Trends 官方端点(真值 0–100,非估算) · 速率: 每分钟几次为宜 搭配阅读: 上一篇讲的收费多源 MCP → Google Search Trends MCP 实战


一、为什么要专门写”免费的那个”

上一篇我们讲了 TrendsMCP——25+ 数据源、交叉验证、很强,但收费(免费额度只有 100 次/月),而且它返回的是”归一化的专有估算值”。

对一个小工作室、或者夫妻店式的小公司来说,100 次/月这个额度,稍微认真做几次多源分析就见底了。这时候你需要一个能天天跑、不心疼的主力雷达。

purahmanian/google-trends-mcp 正好补上这一格。它的定位和收费 MCP 完全互补

维度免费 MCP(本文)收费 TrendsMCP(上一篇)
数据源只有 Google(搜索/热榜/地域)25+ 源(YouTube/TikTok/Reddit/Amazon…)
数据性质Google 官方真值 0–100归一化专有估算值
费用免费(MIT 开源)免费 100 次/月,超了收费
用量悠着点用,几乎无限额度宝贵,要省着用
API Key不需要需要注册
地域下钻interest_by_region
稳定性靠非官方端点,可能被 Google 限流服务商托管,相对稳

一句话总结这套搭配哲学:

免费 MCP 当”广撒网”的日常主力(官方真值、天天跑),收费 MCP 当”精准验证”的关键补刀(多源交叉、省着用)。


二、这个免费 MCP 有什么

它是一个 MIT 开源、TypeScript 写的 MCP Server,直接调 Google Trends 官方端点,不需要任何 API Key。暴露 5 个工具:

工具作用关键参数
interest_over_time最多 5 个词的周级兴趣曲线(0–100)termstimeframe(如 today 5-y
compare_terms2–5 个词归一化对比,直接点出赢家terms
related_queries某个词的热门 + 上升相关搜索term
trending_now指定国家的当日热榜country / geo code
interest_by_region某个词的地域分布(国家或次级地区)term、region scope

注意两个和收费版不一样的关键能力:

  • related_queries 的 “rising”(上升) = 免费拿到 Google 官方的”正在上涨的相关搜索”,这是选题和 breakout 发现的金矿。
  • interest_by_region(地域下钻) = 一个词在哪个省/州/城市最热。这对做本地生意的小公司简直是刚需——收费那个多源 MCP 反而弱在这块。

⚠️ 老实话:它靠的是 Google 非官方端点,Google 随时可能改结构或限流。速率上,每分钟几次比较稳,撞到 HTTP 429 就等 2–5 分钟。它不是企业级 SLA,但对日常捕捉够用。


三、一行装好(真的只要一行)

Claude Code(推荐):

claude mcp add google-trends -- npx -y google-trends-mcp

Claude Desktop: 编辑 claude_desktop_config.json

{
  "mcpServers": {
    "google-trends": {
      "command": "npx",
      "args": ["-y", "google-trends-mcp"]
    }
  }
}

Cursor / Codex:~/.codex/config.toml 里加同样的 npx -y google-trends-mcp

没有 Key、没有注册、没有配额页面。重启客户端,对 AI 说一句就能验证:

“用 google-trends 查一下 ‘City Walk’ 过去 12 个月的兴趣曲线,再看看有哪些 rising 相关搜索。”

如果你把上一篇的收费 MCP 也装了,两个可以同时挂着——AI 会根据你的指令自己选用哪个。这就是双 MCP 搭配的物理基础。


四、小工作室的日常趋势捕捉流水线(双 MCP 版)

下面这套流程,是把两个 MCP 的长处拼起来,专为没有专职数据分析、但需要持续产内容/找选品的小团队设计的。核心原则:能用免费的绝不动收费额度,只在”要下注了”那一刻用收费的做最终确认。

阶段 1 · 广撒网(免费 MCP,天天跑)

每天用免费 MCP 做低成本发现,把额度宝贵的收费 MCP 完全按住不动:

  1. trending_now(你的目标国家)→ 拉当日热榜。
  2. 针对你领域的关键大词,related_queriesrising → 捞正在上涨的长尾。
  3. 对候选词 interest_over_timetoday 3-m / today 12-m)→ 看曲线是不是在起飞(还在往上翘,不是见顶)。

提问模板:

“用 google-trends:① trending_now 拉中国当日热榜;② 对 ‘露营’、‘骑行’、‘飞盘’ 分别取 related_queries 的 rising;③ 把所有正在上升的词用 interest_over_time 查过去 3 个月,只留下曲线还在上涨的。给我一张候选清单。“

阶段 2 · 地域下钻(免费 MCP,本地生意专用)

小公司做的往往是区域生意。用 interest_by_region 看候选词在哪些地区最热:

提问模板:

“用 interest_by_region 查 ‘露营装备’ 在中国各省的兴趣分布,告诉我前 5 个省。如果我要做本地投放/线下活动,应该优先哪几个城市?”

这一步收费 MCP 做不好,是免费 MCP 的独门价值。

阶段 3 · 精准验证(收费 MCP,只在这一步花额度)

阶段 1、2 免费筛出的少数几个真正想下注的候选,才动用收费 MCP 的多源交叉——确认这个需求不是 Google 一家的”一波流”,而是全网(YouTube/TikTok/Reddit/Amazon)同步在涨的真趋势。

提问模板:

“这几个候选词 [A, B],我打算投入做内容/选品。用收费的 trends-mcp 在 YouTube、TikTok、Reddit、Amazon 上分别验证一下:是不是多源同步在涨?只有多源都确认的才值得我下注。”

💡 省钱关键:阶段 1、2 可能查了几十次(全免费),但收费 MCP 只在最后花了 1–2 次。一个月 100 次的免费额度,够你验证几十个真候选——足够小工作室用了。

阶段 4 · 产出分析文章(AI 直接写)

数据都在 AI 上下文里了,直接让它成文:

提问模板:

“基于刚才免费 MCP 的曲线 + 地域数据 + 收费 MCP 的多源验证,帮我写一篇 800 字的趋势分析短文:这个趋势是什么、涨得多快、哪些地区最热、多源是否共振、给我们工作室的 3 条可执行建议(选题/选品/投放)。“


五、把它固化成一个”每日趋势简报”Agent

小工作室最缺的是”每天有人盯趋势”。这套双 MCP 流程可以用 Claude Code 的定时任务固化成一个每天早上自动跑的 Agent:

每天 09:00:
  [免费 MCP]
    1. trending_now → 当日热榜
    2. 我的 3 个领域大词 → related_queries(rising)
    3. 候选词 interest_over_time → 只留起飞中的
    4. interest_by_region → 标注每个候选的热点地区
  [收费 MCP,仅对 Top 3 候选]
    5. 多源交叉验证 → 真趋势 / 一波流
  [产出]
    6. AI 写成《今日趋势简报》:候选 + 地域 + 验证结论 + 建议
  → 推到工作室群 / 我的手机

成本核算:免费 MCP 部分零成本;收费 MCP 每天只碰 Top 3 候选、约 3–9 次请求,一个月约 90–270 次——如果只做工作日、只验证最强候选,免费额度 100 次/月甚至可能刚好够,真需要再升 $19 的 Starter。一个每天帮你盯全网趋势的”分析助理”,成本从 0 到 20 刀封顶。


六、什么时候用哪个:一张决策表

你的需求用哪个 MCP
每天扫热榜、找上升长尾免费(trending_now + related_queries rising)
看一个词是不是在起飞免费(interest_over_time
这个词在哪个地区最热(本地生意)免费独门interest_by_region
要 Google 官方真值、不接受估算免费独门
确认多源共振、避免一波流收费独门(YouTube/TikTok/Reddit/Amazon…)
要下重注前的最终验证收费
追求端点稳定性、企业级收费(托管服务)

七、边界与老实话

  • 只有 Google:免费 MCP 看不到 TikTok/Reddit/Amazon 的独立信号,这正是它需要收费 MCP 补位的原因。
  • 非官方端点:Google 可能限流或改结构,作者说”通常改个 URL 或参数就能修”,但没有 SLA 保证。
  • 速率限制:每分钟几次,别写脚本狂刷;429 就等几分钟。
  • 免费 ≠ 零维护:端点变动时可能要更新 npm 包版本。

但对绝大多数小工作室来说,这套”免费打底 + 收费补刀”的组合,几乎是用最低成本换来一双持续盯全网趋势的眼睛。趋势捕捉从来不是大公司的专利——现在一个两人小团队,用一行 npx 加最多 20 刀/月,就能拥有过去只有专职分析师才做得到的日常趋势雷达。


参考资料

🇬🇧 English

Repo: github.com/purahmanian/google-trends-mcp · MIT · TypeScript Install: claude mcp add google-trends -- npx -y google-trends-mcp · No API key Source: Google Trends official endpoints (real 0–100 values, not estimates) · Rate: a few req/min Companion read: the paid multi-source MCP → Google Search Trends MCP guide


1. Why write up “the free one” specifically

Last time we covered TrendsMCP — 25+ sources, cross-validation, powerful, but paid (free tier is just 100 req/mo), and its values are “normalized proprietary estimates”.

For a small studio or a mom-and-pop company, 100 req/mo runs dry after a few serious multi-source analyses. You need a workhorse radar you can run every day without flinching.

purahmanian/google-trends-mcp fills exactly that slot. It’s perfectly complementary to the paid one:

DimensionFree MCP (this post)Paid TrendsMCP (last post)
SourcesGoogle only (search/board/region)25+ (YouTube/TikTok/Reddit/Amazon…)
Data natureGoogle official ground truth 0–100normalized proprietary estimates
CostFree (MIT)100 free/mo, then paid
UsagePace it, effectively unlimitedPrecious quota, spend sparingly
API keyNoneRequired
Regional drill-downYes (interest_by_region)Weak
StabilityUnofficial endpoints, may be throttledHosted service, steadier

The pairing philosophy in one line:

Free MCP = the daily “wide net” workhorse (official values, run daily); Paid MCP = the “precision confirm” finisher (multi-source, spend sparingly).


2. What the free MCP offers

An MIT-licensed TypeScript MCP server hitting Google Trends’ official endpoints, no API key. Five tools:

ToolFunctionKey params
interest_over_timeWeekly interest (0–100) for up to 5 termsterms, timeframe (e.g. today 5-y)
compare_termsNormalized comparison of 2–5 terms, winner calloutterms
related_queriesTop + rising related searches for a termterm
trending_nowToday’s trending searches for a countrycountry / geo code
interest_by_regionRegional breakdown of interestterm, region scope

Two capabilities that differ from the paid version:

  • related_queries “rising” = Google’s official “rising related searches” for free — a goldmine for topics and breakout discovery.
  • interest_by_region = where a term is hottest by province/state/city. A must-have for local small businesses — and the paid multi-source MCP is actually weak here.

⚠️ Honest note: it rides Google’s unofficial endpoints; Google can change structure or throttle anytime. A few requests per minute is safe; on HTTP 429, wait 2–5 minutes. Not enterprise SLA, but plenty for daily capture.


3. One-line install (really)

Claude Code (recommended):

claude mcp add google-trends -- npx -y google-trends-mcp

Claude Desktop — edit claude_desktop_config.json:

{
  "mcpServers": {
    "google-trends": {
      "command": "npx",
      "args": ["-y", "google-trends-mcp"]
    }
  }
}

Cursor / Codex: add the same npx -y google-trends-mcp to ~/.codex/config.toml.

No key, no signup, no quota page. Restart and verify: “Use google-trends to pull 12 months of interest for ‘City Walk’ and show rising related queries.”

If you also installed the paid MCP from last post, run both at once — the AI picks which to use per your instruction. That’s the physical basis of the dual-MCP pairing.


4. A daily trend-capture pipeline for small studios (dual-MCP)

This stitches both MCPs’ strengths together, built for small teams with no dedicated analyst but a constant need to produce content / find products. Core rule: never touch paid quota when free will do — spend paid only at the moment you’re about to bet.

Stage 1 · Wide net (free MCP, daily)

  1. trending_now (your country) → today’s board.
  2. For your niche’s big terms, related_queries → take rising long-tails.
  3. On candidates, interest_over_time (today 3-m / today 12-m) → is the curve lifting off (still rising, not peaked)?

Prompt:

“Using google-trends: ① trending_now for the US; ② related_queries ‘rising’ for ‘camping’, ‘cycling’, ‘pickleball’; ③ interest_over_time (3-m) on all rising terms, keep only those still climbing. Give me a candidate list.”

Stage 2 · Regional drill-down (free MCP, for local business)

Small companies often run regional businesses. Use interest_by_region:

Prompt:

“Use interest_by_region for ‘camping gear’ across US states; give me the top 5. If I run local ads/events, which cities first?”

This is the free MCP’s exclusive value — the paid one can’t do it well.

Stage 3 · Precision validation (paid MCP, spend quota only here)

Only the few candidates you actually want to bet on get the paid multi-source cross-check — confirming the demand isn’t a Google-only spike but rises across YouTube/TikTok/Reddit/Amazon too.

Prompt:

“Candidates [A, B] — I plan to invest in content/products. Use the paid trends-mcp to validate across YouTube, TikTok, Reddit, Amazon: are they rising in sync? Only multi-source-confirmed ones are worth betting on.”

💡 The money-saver: Stages 1–2 may run dozens of queries (all free), but paid MCP fires only 1–2 times at the end. 100 free/mo covers dozens of real candidates — enough for a small studio.

Stage 4 · Produce the analysis article (AI writes it)

Data’s already in context — let it write:

Prompt:

“From the free-MCP curves + regional data + paid-MCP multi-source validation, write an 800-word trend analysis: what the trend is, how fast it’s rising, hottest regions, whether sources resonate, and 3 actionable moves for our studio (topic / product / ad).“


5. Freeze it into a “Daily Trend Brief” agent

What small studios lack most is “someone watching trends daily.” Freeze this dual-MCP flow into a scheduled Claude Code agent that runs every morning:

Daily 09:00:
  [Free MCP]
    1. trending_now → board
    2. my 3 niche terms → related_queries(rising)
    3. candidates interest_over_time → keep lifting-off only
    4. interest_by_region → tag hot regions per candidate
  [Paid MCP, Top 3 only]
    5. multi-source cross-check → real trend / one-hit spike
  [Output]
    6. AI writes "Today's Trend Brief": candidates + regions + verdict + moves
  → pushed to studio chat / my phone

Cost math: the free part is zero; the paid part touches only Top-3 candidates, ~3–9 req/day, ~90–270/mo — weekdays-only on strongest candidates, the 100 free/mo may even suffice; upgrade to the $19 Starter only if needed. A “daily analyst” watching the whole web, capped at $0–20.


6. Which one when: a decision table

Your needWhich MCP
Daily board scan, find rising long-tailsFree (trending_now + related_queries rising)
Is a term lifting off?Free (interest_over_time)
Where is it hottest (local business)?Free-only (interest_by_region)
Need Google official ground truth, no estimatesFree-only
Confirm multi-source resonance, avoid one-hit spikesPaid-only (YouTube/TikTok/Reddit/Amazon…)
Final check before a big betPaid
Want endpoint stability, enterprise-gradePaid (hosted)

7. Boundaries & honesty

  • Google only: the free MCP can’t see independent TikTok/Reddit/Amazon signals — exactly why the paid one complements it.
  • Unofficial endpoints: Google may throttle or change structure; author says “usually a URL/param tweak fixes it”, but no SLA.
  • Rate limits: a few per minute; don’t hammer it; on 429, wait a few minutes.
  • Free ≠ zero maintenance: you may need to bump the npm package version when endpoints shift.

Still, for most small studios, this “free base + paid finisher” combo buys, at the lowest possible cost, a pair of eyes on the whole web’s trends. Trend capture was never a big-company privilege — now a two-person team, with one npx line plus at most $20/mo, owns a daily trend radar that used to require a full-time analyst.


References


© 2026 Author: Mycelium Protocol. Licensed under CC BY 4.0.

💬 评论与讨论

使用 GitHub 账号登录后发表评论