OpenAbstracts

Log In

Docs

Connect OpenAbstracts to Claude with MCP

Server URL: https://openabstracts.com/mcp

The Model Context Protocol (MCP) lets an AI assistant call external tools. OpenAbstracts runs a remote MCP server, so Claude, Cursor and other MCP clients can search 10M+ academic papers directly in a conversation — no code, no SDK.

What you get

The server exposes one tool, search_papers, with the same parameters as the REST API: a query, an optional mode (papers for titles and abstracts, content for full PDF text), category, year range, journal quartile, and pagination. Results are real bibliographic records with DOIs, abstracts and PDF links, so the assistant can cite sources instead of guessing.

Connect it in Claude

  1. In Claude, open the connectors settings and choose Add custom connector.
  2. Enter the server URL: https://openabstracts.com/mcp
  3. Click Connect. You are taken to an OpenAbstracts page.
  4. Sign in (or create a free account), then choose which of your API keys the connection may use, or create a new one on the spot, and click Allow.
  5. You are sent back to Claude. Ask something like "Find recent papers on supply chain resilience and give me the DOIs."

The connection uses standard OAuth with PKCE. Claude never sees your API key; it receives a short-lived token that resolves to the key you chose, and you can revoke access at any time by revoking that key on the API Keys page.

Other MCP clients

Clients that let you set a static header can skip the OAuth flow and send your key directly:

{
  "mcpServers": {
    "openabstracts": {
      "url": "https://openabstracts.com/mcp",
      "headers": { "x-api-key": "isk_live_YOUR_KEY" }
    }
  }
}

Billing

MCP calls are billed exactly like REST calls: 100 free searches per month, then $1.50 per 1,000 requests from prepaid credit, and searches with no results are never billed.

Make Claude verify citations

Pair the connector with the OpenAbstracts citation Skill, a small instruction file that tells Claude to search for a paper before it cites one and to admit when it can't find it. See why that matters for AI writing tools.