Academia MCP: scientific paper search
Academia MCP
An MCP server for scientific work: search arXiv and ACL Anthology, Semantic Scholar citations, PDF reading and LaTeX compilation
- 94 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The server calls external APIs and downloads files into a workspace directory
- OpenRouter and search service keys live in environment variables
Install
- Works in
Manual install
{"mcpServers":{"academia":{"command":"python3","args":["-m","academia_mcp","--transport","stdio"]}}}Install the package first with `pip3 install academia-mcp`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The server calls external APIs and downloads files into a workspace directory
- OpenRouter and search service keys live in environment variables
This is third-party code. Review the repository files before installing.
What it does
Academia MCP bundles tools for working with scientific papers and datasets in one server. `arxiv_search` and `arxiv_download` search and fetch arXiv papers converted to structured text, `anthology_search` searches the ACL Anthology with date filters and `hf_datasets_search` finds datasets on Hugging Face. `s2_get_citations` and `s2_get_references` show who cites a paper and what it cites, using Semantic Scholar. The server also fetches and normalizes web pages, searches the web through Exa, Brave or Tavily, lists and returns LaTeX templates, `compile_latex` builds a PDF in a workspace directory and `read_pdf` extracts text per page. Some tools use a language model through OpenRouter: `document_qa` answers over a document, PDF review, and research-proposal helpers. The server runs over `stdio`, `sse` and `streamable-http`, with port 5056 by default. HTTP modes can require access tokens via `ENABLE_AUTH`, and the `academia_mcp auth` command issues and revokes them.
Who it is for. Researchers and students who ask Claude Code or Claude Desktop to find papers, trace citations and assemble a LaTeX draft.
Good fit when
- You need recent arXiv papers on a topic with the text in structured form
- You want to see who cites a paper and what it cites
- You look for Hugging Face datasets for a task
- You need to read a PDF page by page and answer questions about it
- You write a paper and want to compile a PDF from LaTeX through the agent
Not a fit when
- You need your personal Zotero library or notes: the server searches open sources
- You have no OpenRouter key: the language-model tools will not work
- You need paywalled journal search: the server cannot reach closed publications
Example request
Find three recent arXiv papers on language model distillation, show who cites them and briefly summarize the main one.MCP
- Transport
- stdio, http, sse
- Authentication
- not required
Environment variables
OPENROUTER_API_KEYsecret- OpenRouter key for language-model tools
EXA_API_KEYsecret- Exa key for web search
BRAVE_API_KEYsecret- Brave key for web search
TAVILY_API_KEYsecret- Tavily key for web search
WORKSPACE_DIR- Directory for PDFs and temporary files
Limitations
Python 3.12 or later is required. The `compile_latex`, `read_pdf`, `download_pdf_paper` and `review_pdf_paper` tools are enabled only when `WORKSPACE_DIR` is set. Language-model tools need `OPENROUTER_API_KEY`, and the unified `web_search` works when at least one of `EXA_API_KEY`, `BRAVE_API_KEY` and `TAVILY_API_KEY` is set. In HTTP mode, access tokens sit in `tokens.json` in plaintext with 600 permissions, so protect the file and use HTTPS. Requests to external APIs (arXiv, Semantic Scholar, search services) leave your network.
How to disable. Remove the `academia` entry from the MCP configuration. The package is removed with `pip uninstall academia-mcp`, and files in `WORKSPACE_DIR` are deleted by hand.
FAQ
How do I connect Academia MCP to Claude Desktop?
Install the package with `pip3 install academia-mcp` and add a server with the command `python3` and the arguments `-m academia_mcp --transport stdio`.
Which sources are supported?
arXiv, the ACL Anthology, Hugging Face Datasets, Semantic Scholar for citations and references, and the regular web through Exa, Brave or Tavily.
Do I need a key?
The README names no key for arXiv and ACL Anthology search. An OpenRouter key is for language-model tools, and an Exa, Brave or Tavily key is for `web_search`.
Can the HTTP server be protected?
Yes. Set `ENABLE_AUTH=true`, issue a token with `academia_mcp auth issue-token` and pass it in an `Authorization: Bearer` header.
How do I build a PDF from LaTeX?
Set `WORKSPACE_DIR`, take a template with `get_latex_template` and call `compile_latex`; the result appears in the workspace directory.
Discussion
No questions or reviews yet.
Related
MCP serversA local MCP server for arXiv literature work: section-level paper reads, BibTeX from metadata, topic watches
Scholar Search MCP connects a discovered paper to its references and later citations when preparing a literature review.
A CLI and MCP server for a research paper library: the agent searches, imports and reads papers and keeps Markdown notes in your vault
MCP server and CLI that searches and downloads books and articles from Anna's Archive inside an agent
BibTeX verification against Crossref, OpenAlex, PubMed and arXiv: a CLI, Python API and MCP server to flag wrong references
A local document base for your agent: hybrid semantic and keyword search plus a web UI for browsing and uploads
README in short
The short README lists features, requirements and two run modes (HTTP and stdio), then describes access tokens and issuing them. Next come a Claude Desktop configuration, a one-line list of tools, environment variables and building and running in Docker.