--- title: "arXiv MCP Server: review, setup & limitations · Undominated.ai" canonical: https://undominated.ai/mcp-servers/arxiv-mcp-server/ description: "Reads arXiv source that has been saved locally and builds BibTeX from arXiv metadata rather than from a PDF scrape." --- # arXiv MCP Server: review, setup & limitations · Undominated.ai > Reads arXiv source that has been saved locally and builds BibTeX from arXiv metadata rather than from a PDF scrape. [← Explore all mcp servers](/mcp-servers/) RESEARCH LITERATURE / blazickjp # arXiv MCP Server Reads arXiv source that has been saved locally and builds BibTeX from arXiv metadata rather than from a PDF scrape. [See setup guidance ↓](#setup)[Original source ↗](https://github.com/blazickjp/arxiv-mcp-server) SOURCE REVIEW Reviewed 2026-10-14 Evidence 3 linked sources Publisher blazickjp Licence [Apache-2.0 ↗](https://github.com/blazickjp/arxiv-mcp-server/blob/eb73c324803e6a270761685052431b42ea9dac34/LICENSE) Revision eb73c324803e [Read what was—and wasn’t—checked ↓](#review) “A local MCP server for agent literature work.” [blazickjp · upstream description ↗](https://github.com/blazickjp/arxiv-mcp-server/blob/eb73c324803e6a270761685052431b42ea9dac34/README.md) Our analysis follows below. 01 / THE REASONING ## Why this made the selection. - The repository description says papers stay on disk and that search is optional. - The README documents `uvx arxiv-mcp-server` as the launch command. - The pinned README is a local arXiv literature server that keeps papers on disk and reads author LaTeX one section at a time. - Search and downloads call external services, but the default transport is local stdio and omitted reads are chunked. ### A good fit for - Search a bounded arXiv query, retain paper/version IDs and read author-submitted LaTeX section by section. - Export BibTeX from arXiv metadata while keeping citation-graph evidence separate from the paper itself. - Reading one section of a downloaded arXiv paper instead of pasting the whole PDF into the client ### Weigh up before choosing - This is a community server, not an arXiv product. arXiv’s API terms still apply to any search the tool performs. - A local copy is not a licence to redistribute the paper. - Citation graphs use the external Semantic Scholar API and cache results locally; they are not an authoritative complete citation census. - Topic watches and downloaded papers create local state. Search/source retrieval/downloads still call external services even though the reading workflow is local. - The Desktop bundle has a separate CPython3.11.x requirement; it is not interchangeable with every uvx environment. - An unrelated npm package uses the same name, so npm or npx is the wrong install. - Omitting max_chars no longer returns the entire paper. - Search is optional and is not the local literature loop. - A pin does not verify transitive dependencies. - This review did not execute the resource. 02 / THE REVIEW RECORD ## What we actually inspected. Source review has boundaries. A clear record is more useful than a “safe” badge. ### Material inspected - README.md - 2026-10-14 reread of retained pinned setup/configuration evidence for the specific operational details added in this batch. - LICENSE ### Our findings - Launch command uvx arxiv-mcp-server is in the README. - Claude and Codex add commands in the same file both point at that uvx invocation. - README documents separate remote search/source/citation services, on-disk watches, storage-path selection and cached-environment refresh. - The server is a local stdio process for arXiv work. Papers stay on disk, and the documented loop is paper id, outline, one section, then citations. - Search, source retrieval, citation graphs, and downloads call external services. The literature loop itself is local. - The supported package is the PyPI name arxiv-mcp-server, not the npm package of the same name. A download or read that omits max_chars returns a 12,000-character chunk rather than the whole paper. ### Not established by this review - No paper was fetched or parsed. - No authenticated session, installation, permission bypass or package-to-source parity test was performed. - The server was not started and its tools were not called. - A pin does not verify transitive dependencies. The review applies to the material and revision named here. A newer upstream release can change its behavior. 03 / PUT IT TO WORK ## Connect a server deliberately. [Upstream setup instructions ↗](https://github.com/blazickjp/arxiv-mcp-server/blob/eb73c324803e6a270761685052431b42ea9dac34/README.md) DOCUMENTED COMMAND uvx arxiv-mcp-server Copy command ↗ Copying does not execute this command. It may retrieve a newer version than the reviewed source. - Launch with uvx, or the client add command that wraps it. - Keep papers on disk. Do not treat a summary as the source. ### Before you start - uv ### Documented configuration example Use the client and authentication context described in the setup guidance. Configuration formats vary between hosts. { "mcpServers": { "arxiv": { "type": "stdio", "command": "uvx", "args": [ "arxiv-mcp-server" ] } } } Copy configuration ↗ ### Compatibility Local stdio MCP clients ### Implementation Python ### Transports stdio ### Local process - Read local paper files - Optional arXiv search if you enable it ### Cost model Apache-2.0 server. arXiv access is subject to arXiv’s terms. 04 / FOLLOW THE EVIDENCE ## The source trail. Our notes are separate from the original resource. Check upstream before adopting a new version. - [Server documentation ↗](https://github.com/blazickjp/arxiv-mcp-server/blob/eb73c324803e6a270761685052431b42ea9dac34/README.md) Checked 2026-10-14 https://github.com/blazickjp/arxiv-mcp-server/blob/eb73c324803e6a270761685052431b42ea9dac34/README.md Supports: Install command and stated tools, summary, install, configExample, access, limitations, bestFor, review, compatibility, transports - [Repository licence ↗](https://github.com/blazickjp/arxiv-mcp-server/blob/eb73c324803e6a270761685052431b42ea9dac34/LICENSE) Checked 2026-10-07 https://github.com/blazickjp/arxiv-mcp-server/blob/eb73c324803e6a270761685052431b42ea9dac34/LICENSE Supports: Redistribution terms - [Server documentation ↗](https://github.com/blazickjp/arxiv-mcp-server/blob/eb73c324803e6a270761685052431b42ea9dac34/README.md) Checked 2026-10-07 https://github.com/blazickjp/arxiv-mcp-server/blob/eb73c324803e6a270761685052431b42ea9dac34/README.md Supports: Install command and stated tools KEEP COMPARING ## Other approaches to consider. Related by category or shared topics. These are alternatives to inspect, not a measured quality order. [### OpenAlex MCP Server ↗ Official OpenAlex server for checking whether a citation exists and for counting scholarly works, with profile-edit tools also registered.](/mcp-servers/openalex/)[### Zotero MCP ↗ Reads the caller’s own Zotero library. It is a community server, not a Zotero product.](/mcp-servers/zotero-mcp/)[### Hugging Face MCP ↗ Connects an assistant to Hub discovery, with optional community tools and account-scoped compute or repository actions.](/mcp-servers/hugging-face/) [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/) ## Continue your investigation - [Inspect resource evidence](/tools/) - [Choose a host](/agents/) - [Inspect reusable workflows](/skills/) - [Use Undominated data](/api/)