Wire Make.com scenarios to live multi-source trends

Trends MCP is the data layer. Make (formerly Integromat) is a visual scenario runner that should call REST, not pretend to be Cursor. The HTTP module posts to https://api.trendsmcp.ai/api with Authorization: Bearer YOUR_API_KEY. Use mode: "get_time_series" for weekly history (MCP name get_trends), mode: "get_growth" for percent windows, and mode: "get_top_trends" for a live feed. Free accounts get 100 requests per month. Starter is $19 per month. Make is the right client when marketers and ops already keep reporting in scenarios, routers, and Data Stores, and the last thing they need is another IDE MCP JSON dialect.

HTTP module fields that actually work

Create an HTTP module. Method POST. URL https://api.trendsmcp.ai/api. Header Authorization with value Bearer YOUR_API_KEY. Body type raw JSON. Example growth call:

{
  "mode": "get_growth",
  "source": "google search",
  "keyword": "air fryer",
  "percent_growth": ["30D", "3M"]
}

A second HTTP module can pull history with "mode": "get_time_series" for the same keyword after the router decides the spike is real. Live discovery uses "mode": "get_top_trends", "type": "Google Trends", "limit": 15.

Do not paste Windsurf serverUrl or Cursor transport into Make. Those keys are editor config. Do not use https://www.trendsmcp.ai/mcp; that is Claude.ai OAuth. Make is Bearer REST (and optionally Bearer MCP if an OpenAI agent module is in the scenario). The docs list every mode. The n8n sibling page describes the same API in node vocabulary if the team runs both tools.

Map JSON with Make’s parser. Routers read results[].growth, direction, recent_date, and baseline_date. Name those mappings once in a Data Store so every scenario does not re-parse by index.

When Make beats n8n, Zapier, and IDEs

Make beats n8n when the operators already think in Make modules, operations quotas, and templates, and do not want to self-host. n8n beats Make when workflows must live as git JSON beside application code. Zapier beats Make when the company already standardized on Zaps and will not buy a second automation seat. None of those products is the trend vendor. Trends MCP is.

IDEs lose on a 06:30 Slack alert. A scenario does not need Cursor Agent to be awake. Interactive debugging of a weird keyword still belongs in Cursor or a REST notebook. Make should stay the delivery rail.

Affiliate, ecommerce, and agency pods already run Make for Sheets and Slack. Giving them the same get_growth math engineering uses in Python removes a second, hand-maintained scraper scenario that drifts every time a DOM class changes.

Routers, filters, and a scenario that stays quiet

Scheduler daily. HTTP get_growth for the campaign keyword list. Filter: 3M growth greater than an agreed number and volume_available true when volume exists. Slack module only then. A second router path writes every result to Google Sheets so silence does not mean “the scenario died.”

Discovery scenarios should not run on Free with get_top_trends across many types. Each type (and each page) is a request. Pick one feed that matches the channel: TikTok Shop Hot Products for shop teams, Amazon Best Sellers Top Rated for marketplace teams, Google Trends for SEO pods.

If an OpenAI module must choose tools, attach MCP at https://api.trendsmcp.ai/mcp with the Bearer header and keep the deterministic HTTP modules for the alert path. Two paths, one key, no OAuth www mix-in.

Operations, Make pricing, and Trends MCP pricing

Make bills operations. Trends MCP bills requests. A scenario can be cheap on Make and expensive on Trends MCP if it fans out sources. 100 free Trends MCP requests per month vanish if a scenario runs 5 keywords, 3 sources, and 30 daily executions. Collapse to weekly, or move to Starter at $19 per month (1,000 requests). Cache in a Make Data Store keyed by source|keyword|mode|date.

Store the API key in Make connections, not in a mapper formula visible in scenario screenshots. Exports of scenarios can leak headers; scrub before sending to a contractor.

Limits of no-code trend ops

Weekly grain means a “hourly spike” scenario is theater. 7D windows help a little and still wobble. Huge percents on tiny baselines will pass a naive greater-than filter; require a minimum baseline_value or volume. invalid_source is usually a Sheet cell that says “Google” instead of google search.

Make is weaker at exploratory charts. Export JSON to Sheets or a warehouse, then chart there. Keep the API as the single measurement core so the Slack number and the sheet number cannot fork.

Agencies that clone a scenario per client should templatize the HTTP module and inject the keyword list from a client-specific Data Store. Cloning the whole scenario including the Bearer header multiplies leak surface. One connection object, many scenarios.

Make operation counts still matter on Make’s own bill. An iterator that posts once per Sheet row is honest. An iterator that also runs a dummy Search module “to be safe” wastes Make operations without changing Trends MCP math. Keep the scenario short: schedule, HTTP, parse, filter, notify, log.

Make is a visual REST client for the shared Trends MCP server. It is not a scraper, not an MCP-first IDE, and not a second data warehouse.