CLI

arxiv <command> [subcommand] [--flags]

arxiv <command> --help is the authority on any command's flags. Every command's long help says what it costs and which plane it lands on, so it is worth reading once even for a command you think you know.

Read

Command	What it does
search [query]	Search arXiv papers
count [query]	Count the results a query has, in one request
paper <id>	Fetch a single paper by arXiv id
author <ref>	Look up an author by name, or by identifier with --id
list <category> [month]	Browse a category listing by month
new <category>	Read today's announcement for a category
categories	List the arXiv categories
category <code>	Show one category
sets	List the OAI-PMH sets
files <ref>	What arXiv serves for a paper
fulltext <ref>	Read the LaTeXML full text
download <ref>	Fetch a PDF, HTML or source to a file
trackbacks [ref]	Inbound links, the external pages that link to a paper

Cite

Command	What it does
bibtex [refs...]	arXiv's own BibTeX entry, or --local to build one from the record
cite [refs...]	bibtex, apa, mla, chicago, ris, csl-json or text, with -s

Graph

Command	What it does
edges <ref>	The claims one read asserts
graph <ref>	Walk the claim graph out from a reference
rdf [ref...]	Write claims as RDF, in Dublin Core and schema.org
crawl [seed...]	Walk arXiv into a store, on a budget
query <sql>	Run read-only SQL over a store
export	Write a store as JSON, NDJSON or CSV
archive <id>...	Write every surface of a paper to disk
db stats	What is in the store, counted three ways
db vacuum	Compact the store

Explain

These six make no network request at all.

Command	What it does
planes	The pace this tool keeps and why
surfaces	The twelve places it reads and what each is for
routes	Every URL it will ever request, with the robots verdict
grammar	arXiv's query language, with examples that work
fields	Every field on a paper, where it comes from and what it costs
predicates [name]	The twenty predicates, with what may be at each end

id <ref> also asks arXiv nothing: it parses a reference locally and prints the canonical id, the style, the DOI, the OAI identifier and the URLs.

Serve

Command	What it does
serve	Serve the operations over HTTP as NDJSON
mcp	Run as an MCP server over stdio
version	Print version information
completion <shell>	Shell completion for bash, zsh, fish or powershell

Flags every command has

Flag	Meaning
-o, --output	auto, table, markdown, list, json, jsonl, csv, tsv, url, raw
--fields	Comma separated columns to show
--no-header	Omit the header row
--template	A Go template applied per record
-n, --limit	Stop after N records, 0 for no limit
--db	Tee every record into a store
--data-dir	Override the data directory
--profile	Named profile to load
--rate	Minimum gap between api plane requests
--html-rate	Minimum gap between arxiv.org requests, floor 15s
--retries	Retry attempts on rate limit or 5xx, -1 for the built-in default
--timeout	Per request timeout
--no-cache	Bypass the on-disk caches
--dry-run	Print actions, do not perform them
--color	auto, always or never
-v, --verbose	Increase verbosity, repeatable
-q, --quiet	Suppress progress output

Flags worth knowing per command

--depth is on paper, edges, rdf, graph and crawl, and takes quick, meta, full or text.

--all is on search and list, and means walk the whole set. On search that gets past the ten thousand result window by slicing on date. On list it pages through a month at the fifteen second pace and says how long that will take before it starts.

--budget is on crawl and graph. On crawl it is paired with --html-budget, because the two planes are five times apart.

--store is on query, export, rdf --from-store and db, and defaults to arxiv.db under the data directory.