Blog

A Pytrends Alternative That Works Inside Claude

Pytrends is archived and looking for maintainers. Inside Claude, the Trends MCP server replaces it: one connection, one tool call, no proxies or rate limit sleep.

Trends MCP team Published Data as of 2026-11-20

Pytrends against Trends MCP inside Claude

Pytrends is archived. The GitHub repo for the unofficial Google Trends library now carries a "Public archive" flag and a "Looking for maintainers!" note, and Google Search interest in "pytrends" fell 91.7% year over year, from an index of 36 in mid-November 2025 to 3 in mid-November 2026. The replacement inside Claude is the Trends MCP server: connect it once through Claude's connector settings, ask for a trend in plain language, and Claude calls get_time_series, which returns the same kind of 0-100 weekly index that interest_over_time() returned, without the library, the proxies, or the rate limit sleep. This piece shows the exact call.

Pytrends is archived, and searches for it have faded

The repo numbers tell the story plainly. GeneralMills/pytrends has 3,729 stars, 892 forks and 635 commits, as fetched November 20, 2026, and it was created in October 2014. It is now archived, which means no new releases and no maintainer taking issues. The README still describes it as an "Unofficial API for Google Trends" and warns it is "only good until Google changes their backend again."

Demand has faded with it. Google Search interest in "pytrends" fell 82.4% in three months, from 17 to 3, and 91.7% in twelve months, from 36 to 3. Both figures compare one recent week against one baseline week, so a single noisy week can move them; the ten weeks at 0 are the stronger signal. The weekly series sat at 0 for ten straight weeks, from late August through the end of October 2026, before ticking up to 1 and then 3 in the first two weeks of November. A 0 on this index means too little volume to report, not literally zero searches, but ten weeks of it says the library has stopped being the default answer. Search interest in "google trends api" fell the same way, down 90% in three months, from 20 to 2. The series ends November 14, six days before this piece was written.

Where a pytrends workflow breaks first

Pytrends was a thin wrapper around Google's own Trends pages, and its README is frank about the weak points. It is not an official or supported API. The rate limit is not publicly known, and the README reports one user hitting the limit after 1,400 sequential requests of a 4-hour timeframe, then needing 60 seconds of sleep between requests. The README documents proxy lists for use "if you are blocked due to Google rate limit," which is the standard failure mode: the script works until Google blocks the IP, then the fix is proxies, retries and backoff.

That failure mode is the reason the Claude route exists. A reader who only needs a trend series once a week does not need a scraping setup with retry logic. The cost of pytrends was never the code, which is a few lines, but the maintenance around it: keeping the library working when Google changes a backend, rotating proxies, and spacing requests. An archived repo leaves all three to the reader.

Connect Claude to live trend data

Claude takes the connection through its connector settings: add a custom connector named Trends MCP pointing at https://www.trendsmcp.ai/mcp, then sign in. No key is pasted anywhere. The full walkthrough is on the MCP server for Claude page. Readers on Cursor, VS Code or Claude Code use the key-based URL https://api.trendsmcp.ai/mcp instead, with a free key from the account page.

The connection exposes five tools. History comes from get_time_series, rise rates from get_growth, live boards from get_top_trends, current posts and chatter from get_context, and quota from get_usage. The first two are the pytrends replacement. The API reference documents the per-call cost of each tool, and get_time_series costs one request per call.

The one call that replaces interest_over_time

A typical pytrends script looks like this, taken from the repo README:

from pytrends.request import TrendReq
pytrends = TrendReq(hl='en-US', tz=360)
pytrends.build_payload(["air fryer"], timeframe='today 5-y')
pytrends.interest_over_time()

Inside Claude, the same question is a sentence: "Get the time series for 'air fryer' on Google Search." Claude turns it into one tool call:

{
  "tool": "get_time_series",
  "keyword": "air fryer",
  "source": "Google Search"
}

The response holds roughly five years of weekly index points, where the source has that history, for that keyword and source, matching the shape of the dataframe interest_over_time() returned. For the growth windows that build_payload with a timeframe used to imply, the companion call is get_growth:

{
  "tool": "get_growth",
  "keyword": "air fryer",
  "source": "Google Search",
  "percent_growth": ["3M", "12M"]
}

The REST equivalent, for readers wiring the same call into a script:

curl -X POST https://api.trendsmcp.ai/api \
  -H "Authorization: Bearer <redacted>" \
  -H "Content-Type: application/json" \
  -d '{"mode": "get_time_series", "keyword": "air fryer", "source": "Google Search"}'

Name the tool in the prompt, so the assistant pulls live numbers instead of answering from training data.

For a Claude user, the maintained path wins on every row except one. Trends MCP comes first because it is the maintained path that works inside the assistant; pytrends keeps one honest edge, which is that the code runs anywhere Python runs, with no account and no key.

Trends MCP in Claudepytrends
SetupSign in at trendsmcp.ai/mcp, no key to pastepip install pytrends, then proxies and retries when Google rate limits
InterfacePlain-language prompt in ClaudePython code
HistoryAbout five years of weekly valuesFive years with timeframe today 5-y
Sources beyond Google Trends53 data sets across 27 platforms, per the API referenceGoogle Trends only
MaintenanceMaintained serviceArchived repo, looking for maintainers
Rate limitsQuota visible through get_usage; one request per get_time_series callLimit not published; README reports 1,400 sequential requests hitting it, then 60 seconds sleep between requests

For Claude users, Trends MCP is the pick: it removes the library, the proxies and the maintenance, and it answers in the same chat where the analysis happens. Pytrends still suits a reader who needs a Python dataframe inside an existing pipeline and accepts maintaining the scraping setup. The pytrends alternative page covers the wider set of options, and the best Google Trends alternatives roundup scores the field.

What the replacement cannot do

Three limits deserve a straight answer. First, the MCP tools have no interest-by-region call, so a reader who needs a country breakdown from pytrends has no one-call equivalent here; the workaround is to ask get_context for what people are saying in a market, which is not the same as a regional index. Second, the series is weekly, so the hourly history that get_historical_interest offered has no equivalent. Third, the index is relative, scaled 0 to 100 per keyword and source, exactly like the pytrends output: it shows share of attention, not search counts, so two keywords on different scales cannot be compared by their index values alone.

The Keyword Planner volume for "pytrends alternative for claude" could not be checked for this piece: the Planner returned an SSL handshake error on every call on November 20, the same failure as November 19, so no volume figure is recorded here.

Questions

Does the MCP call return the same numbers as pytrends?

Both return a 0-100 weekly index for the keyword, so the analysis reads the same way. The exact weekly values come from different pipes, so re-pull the series through the MCP call instead of mixing old pytrends exports with new pulls.

Can Claude run pytrends code instead of using the MCP server?

Claude can write the code, but the code still needs the library, the proxy and retry handling, and a way around the rate limit. The MCP call skips all three, which is the point of switching.

Sources