Paper Search

Paper Search MCP is an open-source MCP server and Claude Code skill for finding academic papers, retrieving lawful open-access PDFs, and extracting paper text within an AI-assisted research workflow.

It searches across arXiv, PubMed, bioRxiv, Semantic Scholar, Crossref, OpenAlex, Europe PMC, and many other research sources through one local server.

Research tasks often move across several databases. You may find a preprint on arXiv, recover a DOI through Crossref, confirm metadata in OpenAlex, and look for an open repository copy before asking an AI assistant to summarize the paper.

Paper Search MCP combines those steps through multi-source search, deduplication, DOI recovery, and open-access fallback retrieval.

Key Features

• Searches multiple academic sources through the unified search_papers workflow.
• Deduplicates overlapping records from different providers.
• Retrieves available open-access PDFs through source-native and repository fallback paths.
• Extracts paper text for AI-assisted reading and synthesis.
• Returns normalized metadata through a shared paper format.
• Runs as a Claude Code skill, MCP server, or local CLI.
• Uses targeted source filters for faster research queries.
• Stores optional provider credentials in environment variables or a .env file.
• Keeps public and open-access sources ahead of restricted services.
• Includes modular connectors for future academic platforms.

Paper Search MCP vs. a Single-Source Search Tool

When a Browser Is Enough

Skip the local setup when your work only needs occasional title lookups, one DOI, or a single arXiv PDF. A browser and the original repository page require less configuration for that job.

Install Paper Search MCP when your research process repeats across several providers and your AI assistant needs structured paper metadata or extracted text. The project has more moving parts than a browser search, and provider reliability still changes by source.

Open-Access Rules and Privacy

Paper Search MCP can locate metadata across many services, but a search result does not guarantee a downloadable PDF. Full-text retrieval depends on publisher-permitted links, public repositories, and source-specific access rules.

Keep routine research on open-access repositories and publisher-approved copies. The project documents an optional restricted-source fallback separately, but it should not form part of a normal academic workflow.

The MCP server runs locally, yet each search query reaches the provider you select. Keep unpublished study titles, participant details, internal research plans, and sensitive project terms out of external provider queries.

How to Install Paper Search MCP

Install uv

Install uv before using the source-clone or uvx paths.

curl -LsSf https://astral.sh/uv/install.sh | sh

Clone and Run the Current Repository

The PyPI package page lists version 0.1.3 from April 29, 2025. The repository documents the broader connector set and current skill workflow, so the clone path gives you the clearest route when a specific connector matters.

git clone https://github.com/openags/paper-search-mcp.git
cd paper-search-mcp
uv run -m paper_search_mcp.server

Install the Claude Code Skill

Create the local Claude Code skill directory.

mkdir -p ~/.claude/skills/paper-search

Copy the skill file from your cloned repository.

cp ~/paper-search-mcp/claude-code/SKILL.md ~/.claude/skills/paper-search/SKILL.md

Replace every <REPO_PATH> entry in ~/.claude/skills/paper-search/SKILL.md with the absolute path to your local clone.

The skill activates when your Claude Code task asks for academic papers, research literature, paper PDFs, or extracted paper text.

Add It to Claude Desktop

Add this block to your Claude Desktop MCP configuration file. Replace /absolute/path/to/paper-search-mcp with your cloned project directory.

{
  "mcpServers": {
    "paper-search-mcp": {
      "command": "uv",
      "args": [
        "run",
        "--directory",
        "/absolute/path/to/paper-search-mcp",
        "-m",
        "paper_search_mcp.server"
      ],
      "env": {
        "PAPER_SEARCH_MCP_UNPAYWALL_EMAIL": "[email protected]",
        "PAPER_SEARCH_MCP_CORE_API_KEY": "",
        "PAPER_SEARCH_MCP_SEMANTIC_SCHOLAR_API_KEY": "",
        "PAPER_SEARCH_MCP_ZENODO_ACCESS_TOKEN": "",
        "PAPER_SEARCH_MCP_GOOGLE_SCHOLAR_PROXY_URL": "",
        "PAPER_SEARCH_MCP_IEEE_API_KEY": "",
        "PAPER_SEARCH_MCP_ACM_API_KEY": ""
      }
    }
  }
}

Claude Desktop stores its MCP configuration in these locations:

Start With Targeted Searches

List available sources before running your first paper search.

uv run --directory ~/paper-search-mcp paper-search sources

Search selected sources first. Targeted searches reduce irrelevant results and avoid unnecessary calls to unstable providers.

uv run --directory ~/paper-search-mcp paper-search search "transformer attention mechanisms" -n 5 -s arxiv,semantic,crossref -y 2020-2026

Read extracted text after you identify a relevant record.

uv run --directory ~/paper-search-mcp paper-search read arxiv 2106.12345

Download a PDF only when the selected source exposes a permitted file.

uv run --directory ~/paper-search-mcp paper-search download arxiv 2106.12345 -o ./downloads

Other Installation Routes

Smithery can add the server to Claude Desktop through one command.

npx -y @smithery/cli install @openags/paper-search-mcp --client claude

uvx runs the package without a permanent installation.

uvx paper-search-mcp

uv tool installs the package as a persistent local tool.

uv tool install paper-search-mcp

pip installs the Python package.

pip install paper-search-mcp

Smithery can also run the project through npx.

npx -y @smithery/cli run @openags/paper-search-mcp

Docker provides an isolated server process.

docker build -t paper-search-mcp .

docker run --rm -i \
  -e [email protected] \
  -e PAPER_SEARCH_MCP_CORE_API_KEY=your_core_key \
  paper-search-mcp

macOS users can avoid uvx wrapper issues by using the clone-and-run path. The repository notes that some uvx setups depend on realpath, which macOS does not include by default.

CLI Commands

uv run --directory <REPO_PATH> paper-search <command> [args]

MCP Workflows

Source Capability Reference

Google Scholar, CiteSeerX, BASE, and SSRN can fail because of provider restrictions, unstable endpoints, or bot detection. IEEE Xplore and ACM Digital Library connectors require keys but still expose skeleton implementations rather than complete search and retrieval support.

Environment Variables

Create a local .env file in the project root when you need provider credentials.

[email protected]
PAPER_SEARCH_MCP_CORE_API_KEY=
PAPER_SEARCH_MCP_SEMANTIC_SCHOLAR_API_KEY=
PAPER_SEARCH_MCP_ZENODO_ACCESS_TOKEN=
PAPER_SEARCH_MCP_GOOGLE_SCHOLAR_PROXY_URL=
PAPER_SEARCH_MCP_IEEE_API_KEY=
PAPER_SEARCH_MCP_ACM_API_KEY=

Set a custom environment-file path when your credentials live outside the repository.

export PAPER_SEARCH_MCP_ENV_FILE=/absolute/path/to/.env

Alternatives and Related Tools

• MCP Servers: Browse MCP servers for research, coding, productivity, and automation workflows.
• Claude Context Local: Add local semantic code search to Claude Code.
• The Ultimate Claude Code Resource List: Find Claude Code skills, MCP servers, plugins, agents, and workflow tools.
• Claude Code Resources: Browse guides, utilities, skills, and companion tools for Claude Code.

Pros

• Free-first source strategy.
• MIT-licensed project.
• Claude Code skill included.
• Broad academic source coverage.
• Open-access fallback retrieval.
• Local CLI access.

Cons

• Local setup takes time.
• Download coverage varies by source.
• Upstream limits remain.
• Google Scholar remains unreliable.
• Some connectors stay incomplete.
• PyPI package may lag.

FAQs

Q: Is Paper Search MCP free?
A: Paper Search MCP uses an MIT license and prioritizes public and open-access sources. Several optional API keys are available at no cost, though some providers apply rate limits or require registration.

Q: Does Paper Search MCP require an API key?
A: Most core searches can run without an API key. Unpaywall requires an email address, while CORE, Semantic Scholar, DOAJ, Zenodo, IEEE, and ACM use optional or source-specific credentials.

Q: Can Paper Search MCP download every academic paper?
A: No. The server can only retrieve PDFs that a selected provider, repository, or publisher link makes available. Metadata-only sources can identify a paper or DOI but cannot guarantee a full-text file.

Q: Is Google Scholar reliable in Paper Search MCP?
A: Google Scholar can return empty results because automated requests often trigger bot detection. Keep it as a secondary discovery path and use public metadata sources such as Crossref, OpenAlex, or Semantic Scholar for routine research.

Q: Does Paper Search MCP keep research queries private?
A: The server process runs locally, but each selected connector sends your query to an external academic provider. Avoid sending confidential research topics or unpublished project details through provider requests.