
arXiv MCP Server (blazickjp): Setup for Claude Code, Cursor and Other Apps
Lets Claude, Cursor and other assistants search arXiv, read papers one section at a time and export citations.
This server connects your assistant to arXiv. It searches papers, downloads them as readable text, reads original LaTeX source, builds BibTeX and follows citations. It can also watch topics and tell you about new papers.
It runs on your machine and needs no account or API key. Papers you download are stored in a folder on your computer. Search, downloads and citation lookups still call arXiv and Semantic Scholar.
This is the version by blazickjp. A separate project by another author uses a similar name.
Good for
- Researchers and students who want an assistant to read papers with them
- Anyone writing a literature review who needs BibTeX and citation links
- Developers tracking new papers in a field with saved topic watches
Works in
12 should work
Tap an app to see how to set it up and any known problems.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Supports local MCP servers.
Client does not support this transport.
Client does not support this transport.
Install arxiv-mcp-server
Claude Code
- 1
Copy this prompt.
Add the arxiv-mcp-server server to Claude Code. Run this command: claude mcp add arxiv -- uvx arxiv-mcp-server Then check that it loaded and tell me if I need to sign in or restart.
- 2
Paste it into Claude Code. It shows the change and asks before running it.
Requires uv (Python).
Built from the package data we hold. Not yet tested by us.
Tools
Searches arXiv for papers by topic, author, category or date
Queries arXiv and returns up to five compact results by default, with title, authors, dates and an abstract snippet. You can sort by relevance or newest first.
Try: “Find recent cs.LG papers on Kolmogorov-Arnold Networks, newest first.”
Network ·
search_papersGets the title and abstract of one paper
Looks up a paper by its arXiv ID and returns its details without downloading the paper.
Try: “Show me the abstract of 2404.19756.”
Network ·
get_abstractSaves a paper to your computer as readable text
Fetches a paper, converts it to Markdown and stores it in a local folder. It returns the first chunk, about 12,000 characters by default.
Try: “Download 2404.19756 so we can read it.”
Write ·
download_paperLists the papers you have saved
Shows the papers already stored in your local folder, with ID, title, authors and date.
Try: “Which papers have I downloaded so far?”
Files ·
list_papersReads a saved paper in chunks
Returns stored text a piece at a time so a long paper does not fill the chat.
Try: “Continue reading 2404.19756 from where we stopped.”
Files ·
read_paperShows a saved paper's section headings
Parses a saved paper into a numbered list of its headings, a page at a time, so you can pick a section to open next.
Try: “Give me the outline of 2404.19756.”
Files ·
get_paper_outlineReads one section of a saved paper
Returns a single section chosen by its outline ID or exact title.
Try: “Read the Methods section of 2404.19756.”
Files ·
read_paper_sectionFinds passages inside a saved paper
Searches the stored text for a phrase and returns matching passages with their positions.
Try: “Find where the paper discusses ablation results.”
Files ·
search_paper_textGets the original LaTeX source of a paper
Downloads the author-submitted source archive from arXiv and returns a limited amount of LaTeX.
Try: “Get the LaTeX source of 2404.19756.”
Network ·
get_paper_latexLists the sections in a paper's LaTeX
Returns a paged outline of the paper's LaTeX source.
Try: “List the LaTeX sections of 2404.19756.”
Network ·
list_paper_latex_sectionsReads one section of a paper's LaTeX
Returns a single LaTeX section chosen by outline ID or exact title.
Try: “Show the LaTeX for the Experiments section.”
Network ·
get_paper_latex_sectionShows what a paper cites and who cites it
Looks up references and citing papers through Semantic Scholar. Results are cached on disk, and a free API key makes it more reliable.
Try: “Who cites 2404.19756, and what does it cite?”
Write ·
citation_graphExports BibTeX for papers
Builds BibTeX entries from arXiv metadata for one or more paper IDs.
Try: “Give me BibTeX entries for 2404.19756 and 1706.03762.”
Network ·
export_citationsSaves a topic to watch for new papers
Stores a standing search, optionally limited to categories, in a local file. Using the same topic again updates it.
Try: “Watch for new papers on diffusion models in cs.CV.”
Write ·
watch_topicLists your saved topic watches
Shows the topics you are watching without checking for new papers.
Try: “What topics am I watching?”
Files ·
list_watchesChecks watched topics for new papers
Returns papers that appeared since the last check, and advances the last-checked time.
Try: “Check my topic watches for anything new.”
Write ·
check_alertsStops watching a topic
Removes a saved topic watch, matched by its exact topic text.
Try: “Stop watching diffusion models.”
Delete ·
unwatch_topicSearches your saved papers by meaning
Finds saved papers similar to a question. The README says it needs the optional pro extras.
Try: “Which of my saved papers discuss attention efficiency?”
Files ·
semantic_searchRebuilds the search index of saved papers
Rebuilds the local index used by semantic search. The README says it needs the optional pro extras.
Try: “Rebuild the index of my saved papers.”
Write ·
reindex
Read from the source code
Permissions
medium risk · 3 things to watch
Medium risk
It reads from arXiv and Semantic Scholar and writes papers and topic watches to a folder on your computer. The only thing it deletes is a saved topic watch you ask it to remove.
What it can do
- Read your data
- Create and change things
- Delete things
- Read-only mode
Watch out for
Saves papers and watches on your disk
Downloaded papers go to ~/.arxiv-mcp-server/papers unless you set another path. Pick a folder you are happy to fill. github.com
It can remove a saved topic watch
The unwatch tool deletes one watch by its exact topic. It does not touch your papers. github.com
Search terms go to arXiv
Queries and paper IDs are sent to arXiv, and citation lookups go to Semantic Scholar. github.com
Security advisories
No published security advisories as of Nov 1, 2026.
Technical details
| Talks to | arxiv.org (search, papers, LaTeX source). Semantic Scholar (citation lookups) |
|---|---|
| Sign-in and keys | No account or API key is needed. An optional free Semantic Scholar key makes citation lookups more reliable. |
Questions
Is this the official arXiv server?
No. It is a community project by blazickjp (the GitHub user), not made by arXiv. It is listed in the official MCP registry, which does not make it an arXiv product.
Does the arXiv MCP server need an API key?
No. It works without any key. The README says an optional free Semantic Scholar key makes citation lookups more reliable.
Is it free?
Yes. It is free and Apache-2.0 licensed. Two tools for searching saved papers by meaning need an optional extra install named pro.
How do I add it to Claude or Cursor?
Use the install box on this page. It needs uv, which provides the uvx command. Do not install it with npm, because an unrelated npm package has the same name.
Where are downloaded papers stored?
In ~/.arxiv-mcp-server/papers by default. Add the storage-path option to the server arguments to use another folder.
Does it keep working offline?
Only for papers you already saved. Search, downloads, LaTeX source and citation lookups call arXiv and Semantic Scholar.
Why does Claude Desktop say it cannot find uvx?
Desktop apps may not see the same PATH as your terminal. The README says to replace uvx in the config with the full path that command -v uvx prints.
Is the tool list complete?
Yes for the current release, which registers 19 tools. The README says two of them, semantic search and reindex, need the optional pro extra.
Information
updated Nov 2, 2026
- Publisher
- Community
- Package
arxiv-mcp-server- Registry
- PyPI
- Transport
- stdio
- Version
- 0.8.2
- License
- Apache-2.0
- Last commit
- Oct 8, 2026
- Open issues
- 4
- Listed in
- Official MCP registry
Sources
Docs, issue trackers and discussions this page is built from. Issues, quotes and findings above link to their own source.
Show all 4Hide
- arxiv-mcp-server README · github.com
- arxiv-mcp-server on PyPI · pypi.org
- arxiv-mcp-server server.py (v0.8.2) · github.com
- arxiv-mcp-server alerts.py (v0.8.2) · github.com