ChatGPT Deep Research excels at multi-step synthesis, but web search alone struggles with dated growth tables and cross-platform comparisons. Connecting Trends MCP supplies structured weekly series, percent-growth windows, and live leaderboards that Deep Research can cite instead of inferring momentum from news articles.
Deep Research was built to browse, compare, and synthesize hundreds of sources into analyst-grade reports. Web search covers news coverage and company filings well. It handles quantitative trend questions poorly: growth percentages, cross-platform comparisons, and dated weekly series rarely appear in crawlable HTML with stable schemas. A connector to Trends MCP fills that gap with JSON the model can quote directly.
For general ChatGPT MCP setup, see MCP server for ChatGPT. For reproducible citation patterns in any LLM report, see grounding LLM research reports with trend JSON.
Deep Research can find articles that mention "GLP-1 interest is rising." It cannot reliably extract a 12-month percent change from Google Search, a TikTok hashtag curve, and Amazon product demand in one pass. News pages omit raw series. Blog posts round numbers. Different outlets use incompatible date ranges.
Structured trend JSON solves three problems at once:
date and a normalized value from 0 to 100.get_growth(keyword='GLP-1', source='google search, tiktok', percent_growth=['12M']) and expect the same schema.The data shows what changed. Web search explains why. Deep Research needs both.
ChatGPT Settings exposes custom connectors (formerly called connectors) under Apps and Connectors. The remote MCP URL is:
https://api.trendsmcp.ai/mcp
Authentication uses a Bearer token with the API key from trendsmcp.ai/account. Transport is HTTP (streamable HTTP). After saving, start a Deep Research task and include routing language in the prompt so the agent selects Trends MCP instead of default web search.
Example Deep Research prompt:
Using TrendsMCP, compare 12M Google Search growth for "AI agents" versus "MCP server".
Then use web search for product launch news that aligns with the inflection dates in the series.
Cite the exact growth percentages and recent_date fields from Trends MCP in the report.
The using TrendsMCP prefix matches the routing pattern documented on trendsmcp.ai/docs.
When the question asks how fast something grew, get_growth returns preset windows in one call:
{
"source": "google search",
"keyword": "semaglutide",
"percent_growth": ["12M", "3M", "YTD"]
}
Response fields include growth, direction, recent_date, baseline_date, and volume_growth where volume is available. Deep Research should quote those fields in the body text, not paraphrase them.
When the question asks when interest inflected, get_trends returns roughly five years of weekly points:
{
"source": "google search",
"keyword": "semaglutide",
"data_mode": "weekly"
}
The model can align spikes in the series with dated news events found through web search.
When the question asks what is trending right now, get_top_trends returns ranked leaderboards:
{
"mode": "top_trends",
"type": "Google Trends",
"limit": 25
}
Feed names are case-sensitive. Valid values include Google Trends, TikTok Trending Hashtags, Reddit Hot Posts, Wikipedia Trending, and X (Twitter).
Using TrendsMCP, pull 6M growth for [Brand A] and [Brand B] on google search, youtube, and tiktok.
Then run Deep Research on positioning changes, pricing moves, and retail partnerships for both brands since [start date].
Require a table that pairs each Trends MCP growth figure with at least one dated news source.
Using TrendsMCP, chart weekly google search interest for [product name] for the past 18 months.
Mark the launch week. Then search the web for review coverage and channel conflict stories in the six weeks after launch.
Explain whether search interest led or lagged press coverage.
Using TrendsMCP, fetch get_top_trends for Google Trends and Wikipedia Trending (limit 15 each).
Then research the top three overlapping topics for corporate announcements in the past 48 hours.
Distinguish topics with sustained 12M growth (get_growth) from one-day spikes.
OpenAI exposes Deep Research through the Responses API with o3-deep-research and o4-mini-deep-research. Those models accept web search, code interpreter, file search, and remote MCP servers that implement search and fetch tools per OpenAI's company-knowledge compatibility schema.
Trends MCP exposes get_trends, get_growth, and get_top_trends. That schema does not match the search/fetch interface Deep Research API models require. Practical options:
| Approach | Model | Trends data path |
|---|---|---|
| ChatGPT UI Deep Research | o3-deep-research (chat) | Custom MCP connector with standard tools |
| Responses API agent | o3 or gpt-4.1 | type: mcp block with server_url: https://api.trendsmcp.ai/mcp |
| Responses API deep research | o3-deep-research | Pre-fetch JSON, attach via file search vector store |
| Batch pipeline | Any | REST POST to https://api.trendsmcp.ai/api, inject results into prompt |
For a full Responses API wiring guide with Python and curl examples, see OpenAI Responses API with Trends MCP.
Deep Research decomposes prompts into sub-questions. A single user request might trigger three get_growth calls (different keywords), two get_trends series, and one live feed pull. That is six Trends MCP requests.
Public Trends MCP plans as of July 2026:
| Plan | Monthly price | Included requests |
|---|---|---|
| Free | $0 | 100 |
| Starter | $19 | 1,000 |
| Pro | $49 | 5,000 |
| Business | $199 | 25,000 |
Narrow the prompt when testing on Free tier. Specify allowed sources in the prompt (google search only) instead of letting the agent query five platforms per keyword.
Trends MCP does not replace filings, earnings transcripts, clinical papers, or forum threads. It answers how loud a topic is and whether momentum is accelerating. Deep Research should still browse for mechanism, regulation, and competitive response.
The strongest reports pair one or two quantitative Trends MCP calls with web search on the same entities. A sentence that cites 12M growth: +34.2% (recent_date: 2026-06-21) next to a dated FDA headline reads as research. A sentence that says "interest seems up" reads as filler.
For live-feed architecture beyond ChatGPT, see real-time trends API. For LLM workflow patterns outside OpenAI, see LLM-native trend research workflow.
FAQ