Answers backed by research papers, not the model's memory.

When a claim needs evidence, it narrows the search to study types such as meta-analyses and randomized trials, by year and sample size, and cites each paper with a link.

What it is

A remote MCP server Consensus runs at mcp.consensus.app/mcp. It gives your assistant one tool: a search across peer-reviewed research papers, which Consensus puts at over 400 million. It works in Claude, ChatGPT (including Deep Research), Microsoft 365 Copilot, Gemini Spark, Cursor and other clients.

What you get

Papers found for a research question, with titles, journals, years, citation counts and links
Results narrowed by publication year, study type (such as randomized trials, meta-analyses and systematic reviews), journal quality and sample size
In ChatGPT Deep Research: repeated searches turned into a cited summary

Requirements

A free Consensus account is enough to sign in; with no account it still works, at reduced limits.
Clients without sign-in support, and your own scripts, can send a Consensus API key instead.
Your Consensus plan, not your AI subscription, sets how many papers each search returns and how many searches you get a month.
If every search returns only 3 papers, your API key is wrong or revoked; the server falls back to no-account limits instead of showing an error.

Setup effort

Paste a URL with a key — add the endpoint to your client