MCP Server for E-Commerce Data: Marketplaces, Reviews and Prices as Agent Tools

It's a Model Context Protocol server, published as the npm package @roketfy/data-engine-mcp, that exposes every endpoint on the Data Engine — Trendyol, Hepsiburada, Amazon, n11, Akakçe, Google and the rest — as a separate tool an MCP-enabled agent can call. The tools are generated from the same registry that powers the public HTTP API and its documentation, so nothing is hand-wired or falls behind: an endpoint added to the API becomes a tool here automatically, with the identical input schema the HTTP API itself validates against.

Calls made through the MCP server go out over the same public API your own code would call, under your own token — the same plan limits, the same concurrency ceiling, the same per-call credit cost, and every call still lands in your dashboard's request log exactly like a direct HTTP request would.

Setup is one command plus an environment variable. In Claude Code, you can add the server from a terminal, pointing it at the npx package and passing your API token as DATA_ENGINE_API_TOKEN — the token itself lives in the sidebar of your dashboard. Claude Desktop, Cursor and other MCP clients that read the standard config shape can instead have the server block added directly to their own configuration file: a command of npx, arguments of -y and the package name, and the same DATA_ENGINE_API_TOKEN environment variable.

There's no separate signup for the MCP server itself — it's the same account and the same token you'd use to call the HTTP API directly, so switching between calling an endpoint from your own code and calling it from an agent doesn't require managing two credentials.

All of them. Every endpoint documented on the platform — marketplace listings, product reviews and questions, price history, search results, news, trends, Maps and review-site data — has a matching MCP tool, named after its platform and slug (for example, trendyol_reviews or amazon_product_detail). The six endpoints in the table above are a representative slice covering marketplace detail, reviews, location reviews, general search and price history, but they're a sample, not the ceiling: the full catalog is as large as the HTTP API itself, and it's documented over MCP too, so an agent can read the docs directly rather than needing a human to point it at a web page first.

Because tools are generated from the same source as the API, the input fields an agent has to supply — things like a product URL, a contentId, a place keyword or a page number — match exactly what the corresponding HTTP endpoint's documentation describes.

A few examples of the kind of request that becomes a direct tool call rather than a manual copy-paste job: 'pull the reviews for this Trendyol product and summarize the recurring complaints,' 'compare this item's Akakçe price history against its current Trendyol listing and tell me if the price today is unusually high,' or 'look up this business's Google Maps reviews and tell me its most common negative theme.' In each case the agent calls the matching tool directly — trendyol_reviews, akakce_price_history, google_maps_reviews — reads the structured JSON that comes back, and reasons over it in the same turn.

Because the same registry backs both the HTTP API and the MCP tools, an agent workflow built this way stays in sync with the platform automatically — a new endpoint shows up as a new tool without any change to the agent's configuration.

No. Credits are shared across both access paths — a call an agent makes through a tool is charged exactly the same as the identical call made directly against the HTTP endpoint, at the credit cost listed for that endpoint, and only on success. There's no separate MCP pricing tier and no extra fee for agent access; it draws from the same one-time credit package as everything else on the platform.

Yes. Every MCP tool call is a real request against the public API under your token, so it shows up in your dashboard's request log the same way a call from your own code would — with the endpoint, the parameters sent and the credits charged. That gives you an audit trail for what an agent did, without having to instrument the agent itself to log it.