Niched exposes its evidence over the Model Context Protocol, so Claude, Cursor, or any MCP client can query it while you are choosing software. Read only. An MCP client has no browser session, so it identifies itself with a free access token - that is what lets Niched show you your own query history.
Include the trailing slash. The transport is Streamable HTTP, so requests are POSTed withAccept: application/json, text/event-stream. Opening the URL in a browser will not work and is not meant to.
Available tools
Facts, and the shape of what is missing.
01list_categories
Every category and how many products sit in each.
02get_category
The products in one category, with the evidence held for each and the gaps that remain.
03compare_adoption
Official SDK downloads compared inside one category, from dated registry windows.
04get_tool
One product: identity, categories, evidence coverage, and what is missing.
05recent_activity
What entered the catalog and what was re-captured, by day.
What it will not do
There is no recommendation tool.
Niched has not published a reviewed niche verdict yet, so the server cannot return one. Adoption is observed integration, not a ranking: the most-downloaded product in a category is not therefore the right one for your constraints. Every response carries that limit in its own payload so a model repeating it has to repeat the caveat too.
What it will do
Return dated evidence and name the gaps.
Every measurement carries a capture date and its source. Where Niched holds nothing, the response says so rather than returning a zero. Pricing, buyer outcomes, and satisfaction are currently unmeasured for most products, and the server reports that directly.