semantic-scholar-rs

Unofficial Rust SDK and MCP server for the Semantic Scholar API.

Note: This is a community-maintained project and is not affiliated with or endorsed by the Semantic Scholar team at the Allen Institute for AI.

Features

Paper search, lookup by ID / DOI / ArXiv ID
Citations and references
Author search and lookup
Paper recommendations
Automatic retry with exponential backoff (429 / 5xx)
Optional MCP server for LLM tool-use integration

Quick Start

[dependencies]
semantic-scholar-rs = "0.1"
tokio = { version = "1", features = ["full"] }

use semantic_scholar::SemanticScholar;

#[tokio::main]
async fn main() -> semantic_scholar::Result<()> {
    let client = SemanticScholar::new()?;

    // Search papers
    let results = client.search_papers("attention is all you need")
        .limit(5)
        .send()
        .await?;

    for paper in &results.data {
        println!("{paper}");
    }

    // Get a specific paper
    let paper = client.get_paper("204e3073870fae3d05bcbc2f6a8e263d9b72e776")
        .send()
        .await?;
    println!("{paper}");

    Ok(())
}

API Key

Unauthenticated requests share a global rate-limit pool (100 req/min across all unauthenticated users). For dedicated limits, obtain a key from Semantic Scholar API and pass it to the client:

let client = SemanticScholar::with_api_key("your-key")?;

Endpoints

Method	Description
`search_papers(query)`	Keyword search with filters (year, field of study, etc.)
`get_paper(id)`	Lookup by S2 ID, `DOI:...`, `ARXIV:...`, `CorpusId:...`
`get_paper_by_doi(doi)`	Convenience wrapper for DOI lookup
`get_paper_by_arxiv(id)`	Convenience wrapper for ArXiv lookup
`get_citations(paper_id)`	Papers that cite the given paper
`get_references(paper_id)`	Papers referenced by the given paper
`search_authors(query)`	Author name search
`get_author(author_id)`	Author details by S2 author ID
`get_recommendations(paper_id)`	Recommended papers based on a paper

All request builders support .limit(), .offset(), and .fields() for pagination and field selection.

MCP Server

An MCP server is included as an optional binary, exposing all endpoints as LLM-callable tools over stdio.

Installation

# Homebrew (macOS/Linux)
brew install ynishi/semantic-scholar/semantic-scholar-mcp

# Shell script (macOS/Linux)
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/ynishi/semantic-scholar-rs/releases/latest/download/semantic-scholar-rs-installer.sh | sh

# cargo install (requires Rust toolchain)
cargo install semantic-scholar-rs --features mcp

Build from Source

cargo build --release --features mcp
./target/release/semantic-scholar-mcp

Claude Desktop Configuration

{
  "mcpServers": {
    "semantic-scholar": {
      "command": "/path/to/semantic-scholar-mcp",
      "env": {
        "SEMANTIC_SCHOLAR_API_KEY": "your-key"
      }
    }
  }
}

Available Tools

Tool	Description
`search_papers`	Search academic papers by keyword
`get_paper`	Get paper details by identifier
`search_authors`	Search authors by name
`get_author`	Get author details by ID
`get_citations`	Get citing papers
`get_references`	Get referenced papers
`get_recommendations`	Get paper recommendations

Testing

# Unit tests (offline, CI-safe)
cargo test

# MCP protocol E2E (no API calls)
cargo test --features mcp --test mcp_e2e_test

# Full integration tests (hits real API)
cargo test --features mcp,integration-test -- --test-threads=1

License

MIT

semantic-scholar-rs 0.1.1