Product-Research MCP Suite: giving AI assistants hands on market data

The problem

Researching products for Puzzably and AlloyHaus meant juggling Keepa charts, Google Trends tabs, and Jungle Scout dashboards, then synthesizing it all by hand. AI assistants are excellent at the synthesis but had no hands on the data. The Model Context Protocol fixes exactly that, so I built the missing servers.

What I built

Three standalone open-source MCP servers, each installable with one npx command in Claude Desktop, Claude Code, or Codex:

• keepa-mcp: Amazon price and sales-rank history via the Keepa API. Six tools covering product lookups, price history, rank trends, search, best sellers, and deals.
• google-trends-mcp: demand signals from Google Trends with no API key. Five tools for interest over time, term comparison, related queries, trending searches, and regional interest.
• junglescout-mcp: Amazon keyword volumes, product database queries, sales estimates, and brand share of voice via the Jungle Scout API. Five tools.

All three share one architecture: TypeScript, stdio transport, zod-validated inputs, helpful error text instead of thrown exceptions, and fully mocked test suites (82 tests across the suite) so CI never makes a live call.

The hard part: unofficial and underdocumented APIs

Mocked tests prove your logic; they cannot prove the upstream contract. Live smoke testing caught what mocks never could:

• Google Trends had quietly changed everything: cookie-less requests now get rejected, one endpoint was removed outright, and the JSON anti-hijacking prefix differs between endpoints. The server now bootstraps a session cookie from the rejection response and falls back to the RSS feed that replaced the dead endpoint.
• The Jungle Scout API returns 404 for every request unless you send a versioned Accept header that is easy to miss in the docs. The original client had the wrong header, wrong HTTP methods, and wrong response field names. I rebuilt it against the official OpenAPI spec and verified every tool against the live API.

The payoff

The suite did real work immediately: an anchored Google Trends screen across twenty product niches surfaced a breakout category, and Jungle Scout data verified listings doing over a million dollars a month in it. That research picked the next product on this page.

The model

Free and open source, MIT licensed, distributed through npm and the MCP registries. No hosting, no telemetry, no paid tiers. It earns through affiliate referrals in the READMEs and as proof of work for custom MCP builds.