home / mcp / scientific papers mcp server
Provides real-time access to scientific papers from arXiv, OpenAlex, PMC, Europe PMC, bioRxiv/medRxiv, and CORE via MCP protocol and CLI.
Configuration
View docs{
"mcpServers": {
"benedict2310-scientific-papers-mcp": {
"command": "npx",
"args": [
"-y",
"@futurelab-studio/latest-science-mcp@latest"
],
"env": {
"CORE_API_KEY": "YOUR_API_KEY"
}
}
}
}You can access a wide range of scientific papers in real time through a dedicated MCP server that combines six major sources. It provides convenient, programmatic ways to discover, fetch, and read papers from arXiv, OpenAlex, PubMed Central, Europe PMC, bioRxiv/medRxiv, and CORE, all via a single interface. This enables you to search by topic or author, retrieve metadata, and obtain full-text content when available, with smart fallbacks and rate-limiting to keep usage reliable.
You will run the MCP server locally or connect to its public endpoint using your MCP client. Once connected, you can discover available paper categories from all sources, fetch the latest papers by category, search across sources by keywords or authors, and request full-text content for any paper by ID. The server returns structured paper metadata and the full text when extraction succeeds, with graceful fallbacks if text extraction is unavailable.
Prerequisites you need before installation: Node.js (preferred recent LTS) and npm. You may also want a shell environment for running commands.
Install dependencies and build the MCP server package, then start using one of the available runtime options.
npm install
npm run build
```
To run via npx (recommended for AI tools):
```
```
Or install globally and run the CLI binary if you prefer:
```
npm install -g @futurelab-studio/latest-science-mcp
```
Then configure your MCP client to point at the local server or the remote endpoint as described in the connection guidance.The server supports multiple connection modes. If you want to connect via a remote HTTP endpoint, configure the HTTP method with the server URL that serves the MCP API. If you want to run a local MCP process, you can start a server using the CLI commands described below.
Configure access control and API keys if your deployment requires restricted usage. When using external services like CORE, you may need separate API keys; keep these credentials secure and avoid exposing them in client configurations.
If you encounter rate limiting, try reducing the requested count, spreading requests over time, or using caching where available. For text extraction failures, the system will return metadata and a partial text excerpt when possible, with warnings indicating any missing content.
This MCP server exposes a range of capabilities to manage and explore scientific papers across all six sources. You can list categories, fetch the latest papers, search by keywords or authors, fetch top-cited papers since a date, and fetch full content for a particular paper by its ID. The tools work with a unified data model that includes paper id, title, authors, date, pdf_url, and the extracted text.
Lists available categories/concepts from any data source. Returns an array of category objects with id, name, and optional description.
Fetches the latest papers from any source for a given category with metadata only (no text extraction). Returns an array of papers with id, title, authors, date, pdf_url, and text set to an empty string.
Fetches the top cited papers from OpenAlex for a given concept since a specified date. Returns papers with metadata and citation information.
Searches for papers across multiple sources with field-specific queries and sorting options. Returns papers with metadata and an empty text field for later fetch_content calls.
Fetches full metadata and complete text content for a specific paper by ID, performing full-text extraction and returning the complete text content with metadata.