yarxiv_mcp

MCP server with tools to search, fetch, analyze, and report on scientific papers and datasets.

• ArXiv search and download
• ACL Anthology search
• Hugging Face datasets search
• Semantic Scholar citations and references
• Web search via Exa, Brave, or Tavily
• Web page crawler, LaTeX compilation, PDF reading
• Optional LLM-powered tools for document QA and research proposal workflows

• Python 3.12+

• Using pip (end users):

pip3 install academia-mcp

• For development (uv + Makefile):

uv venv .venv
make install

• Run over HTTP (default transport):

python -m academia_mcp --transport streamable-http
# OR
uv run -m academia_mcp --transport streamable-http

• Run over stdio (for local MCP clients like Claude Desktop):

python -m academia_mcp --transport stdio
# OR
uv run -m academia_mcp --transport stdio

Notes:

• Transports: stdio, sse, streamable-http.
• host/port are used for HTTP transports; ignored for stdio. Default port is 5056 (or PORT).

Academia MCP supports optional token-based authentication for HTTP transports (streamable-http and sse). Authentication is disabled by default to maintain backward compatibility.

Set the ENABLE_AUTH environment variable to true:

export ENABLE_AUTH=true
export TOKENS_FILE=/path/to/tokens.json  # Optional, defaults to ./tokens.json

Issue a new token:

academia_mcp auth issue-token --client-id=my-client --description="Production API client"

# Issue token with 30-day expiration
academia_mcp auth issue-token --client-id=test-client --expires-days=30

# Issue token with custom scopes
academia_mcp auth issue-token --client-id=admin --scopes="read,write,admin"

List active tokens:

academia_mcp auth list-tokens

Revoke a token:

academia_mcp auth revoke-token mcp_a1b2c3d4e5f6...

Include the token in the Authorization header with the Bearer scheme or as a query parameter apiKey.

Security Notes:

• Tokens are displayed only once during issuance. Store them securely.
• Use HTTPS in production to protect tokens in transit.
• The tokens.json file is automatically created with restrictive permissions (mode 600).
• Tokens are stored in plaintext (standard practice for bearer tokens) - protect the tokens file.

{
  "mcpServers": {
    "academia": {
      "command": "python3",
      "args": [
        "-m",
        "academia_mcp",
        "--transport",
        "stdio"
      ]
    }
  }
}

• arxiv_search: Query arXiv with field-specific queries and filters.
• arxiv_download: Fetch a paper by ID and convert to structured text (HTML/PDF modes).
• anthology_search: Search ACL Anthology with fielded queries and optional date filtering.
• hf_datasets_search: Find Hugging Face datasets with filters and sorting.
• s2_get_citations: List papers citing a given arXiv paper (Semantic Scholar Graph).
• s2_get_references: List papers referenced by a given arXiv paper.
• visit_webpage: Fetch and normalize a web page.
• web_search: Unified search wrapper; available when at least one of Exa/Brave/Tavily keys is set.
• exa_web_search, brave_web_search, tavily_web_search: Provider-specific search.
• get_latex_templates_list, get_latex_template: Enumerate and fetch built-in LaTeX templates.
• compile_latex: Compile LaTeX to PDF in WORKSPACE_DIR.
• read_pdf: Extract text per page from a PDF.
• download_pdf_paper, review_pdf_paper: Download and optionally review PDFs (requires LLM + workspace).
• document_qa: Answer questions over provided document chunks (requires LLM).
• extract_bitflip_info, generate_research_proposals, score_research_proposals: Research proposal helpers (requires LLM).

Availability notes:

• Set WORKSPACE_DIR to enable compile_latex, read_pdf, download_pdf_paper, and review_pdf_paper.
• Set OPENROUTER_API_KEY to enable LLM tools (document_qa, review_pdf_paper, and bitflip tools).
• Set one or more of EXA_API_KEY, BRAVE_API_KEY, TAVILY_API_KEY to enable web_search and provider tools.

Set as needed, depending on which tools you use:

• OPENROUTER_API_KEY: required for LLM-related tools.
• BASE_URL: override OpenRouter base URL.
• DOCUMENT_QA_MODEL_NAME: override default model for document_qa.
• BITFLIP_MODEL_NAME: override default model for bitflip tools.
• TAVILY_API_KEY: enables Tavily in web_search.
• EXA_API_KEY: enables Exa in web_search and visit_webpage.
• BRAVE_API_KEY: enables Brave in web_search.
• WORKSPACE_DIR: directory for generated files (PDFs, temp artifacts).
• PORT: HTTP port (default 5056).

You can put these in a .env file in the project root.

Build the image:

docker build -t academia_mcp .

Run the server (HTTP):

docker run --rm -p 5056:5056 \
  -e PORT=5056 \
  -e OPENROUTER_API_KEY=your_key_here \
  -e WORKSPACE_DIR=/workspace \
  -v "$PWD/workdir:/workspace" \
  academia_mcp

Or use existing image: phoenix120/academia_mcp

• Comprehensive report screencast (YouTube)
• Single paper screencast (YouTube)

• make install: install the package in editable mode with uv
• make validate: run black, flake8, and mypy (strict)
• make test: run the test suite with pytest
• make publish: build and publish using uv

Only needed for LaTeX/PDF tools. Ensure a LaTeX distribution is installed and pdflatex is on PATH, as well as latexmk. On Debian/Ubuntu:

sudo apt install texlive-latex-base texlive-fonts-recommended texlive-latex-extra texlive-science latexmk