Consensus MCP Server

Consensus is an AI-powered academic search engine that indexes over 200 million peer-reviewed research papers from sources like Semantic Scholar, PubMed, and arXiv. The official Consensus MCP server lets AI clients (Claude Desktop, ChatGPT, Cursor, Claude Code, Codex, Windsurf, VS Code) query that corpus directly inside a conversation and get cited, evidence-backed answers without leaving the chat.

The server exposes two tools. A search tool accepts a natural-language research question along with structured filters (publication year range, study design, journal SJR quartile, human-only studies, minimum sample size, medical mode, preprint exclusion, study duration). A fetch tool (currently ChatGPT Deep Research only) retrieves the full details of a paper by its Consensus ID, including title, abstract, authors, journal, year, citation count, and a direct link.

Consensus is provider-hosted at https://mcp.consensus.app/mcp and supports OAuth for end users plus Bearer-token authentication for enterprise API access. Access tiers range from a no-account option (3 papers per search) up to Enterprise (unlimited with API key), so teams can integrate it into agentic research workflows without managing infrastructure.

Tools

Consensus is a remote provider-hosted MCP server at https://mcp.consensus.app/mcp. No install is required; you connect your MCP client to the URL and authenticate via OAuth.

Prerequisites

• An MCP-compatible client (Claude Desktop, ChatGPT, Claude Code, Codex, Cursor, Windsurf, VS Code)
• A Consensus account is optional (a no-account tier exists); free and paid plans raise per-search and monthly limits

Claude Desktop Settings → Connectors → Browse → search "Consensus" → Connect, then sign in via OAuth.

ChatGPT Apps sidebar → search "Consensus" → Connect.

Claude Code

claude mcp add --transport http consensus https://mcp.consensus.app/mcp

Codex

codex mcp add consensus --url https://mcp.consensus.app/mcp
codex mcp login consensus

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "Consensus": {
      "command": "npx",
      "args": ["-y", "mcp-remote@latest", "https://mcp.consensus.app/mcp"]
    }
  }
}

VS Code (GitHub Copilot)

{
  "servers": {
    "Consensus": {
      "url": "https://mcp.consensus.app/mcp"
    }
  },
  "inputs": []
}

Windsurf

{
  "mcpServers": {
    "consensus": {
      "serverUrl": "https://mcp.consensus.app/mcp"
    }
  }
}

Enterprise / API key For programmatic access, send Authorization: Bearer YOUR_API_KEY to https://mcp.consensus.app/mcp instead of using OAuth.

• Run an evidence-based literature review filtered to RCTs and meta-analyses published in the last five years in top-quartile journals.
• Pull human-only clinical studies with a minimum sample size to answer a medical question inside Claude or ChatGPT.
• Draft a grant background section with inline citations by chaining search queries on multiple sub-topics.
• Have ChatGPT Deep Research fetch full abstracts and metadata for a shortlist of papers before summarizing them.
• Build an agent that screens new arXiv/PubMed publications on a topic each week and surfaces high-quality ones, excluding preprints.

• "Find meta-analyses from 2020 onward on the effect of GLP-1 agonists on cardiovascular outcomes, human studies only."
• "Search Consensus for randomized controlled trials on mindfulness for chronic pain with sample size over 200."
• "Give me top-quartile journal papers on transformer architectures published in 2024, excluding preprints."
• "Pull a literature summary on microplastics and gut health and cite each claim with the Consensus paper link."
• "Using Deep Research, fetch full details for these five Consensus paper IDs and compare their methodologies."

• Official, provider-hosted server maintained by Consensus, so no self-hosting or key management required for OAuth users.
• Rich, research-specific filters (study type, SJR quartile, human-only, sample size, medical mode) that go well beyond keyword search.
• Broad client support: Claude Desktop, ChatGPT, Claude Code, Codex, Cursor, Windsurf, and VS Code.
• Usable without a paid Claude or Consensus plan thanks to the no-account and free tiers.

• Per-search and monthly query limits on free tiers; serious research workloads require a Pro, Teams, or Enterprise plan.
• The fetch tool is currently limited to ChatGPT Deep Research, so other clients only get search-result snippets.
• Coverage and metadata depth (DOIs, key takeaways) vary by plan; full enterprise metadata requires an API key.

• PubMed MCP servers (community implementations) for NIH-indexed biomedical literature only.
• Semantic Scholar API wrapped as an MCP server for raw academic search without Consensus filters.
• arXiv MCP server for preprint-focused search in CS, physics, and math.