Arxiv

# ArXiv Plugin Search, retrieve, and download academic papers from ArXiv directly in Claude Code. ## Features - **Paper search** — query by title, author, abstract, category with Boolean operators - **Paper metadata** — full details: authors, abstract, categories, DOI, links - **PDF download** — save papers locally for reading or indexing ## Quick Start ``` /plugin marketplace add brunowinter8192/claude-plugins /plugin install arxiv ``` Restart the session after installation. The plugin is ready to use — no API keys required. ## Prerequisites - Python >= 3.10 - Internet connection (ArXiv API is public and free, no auth required) ## Setup For manual installation without the plugin marketplace: ```bash git clone https://github.com/brunowinter8192/Arxiv.git cd Arxiv python -m venv .venv ./.venv/bin/pip install -r requirements.txt ``` The plugin installs an `arxiv-cli` wrapper at `~/.local/bin/arxiv-cli`. No API keys required. ## Usage ### CLI Subcommands (via `arxiv-cli`) The `arxiv-search` Skill calls these automatically via the `arxiv-cli` wrapper. You can also run them directly from the terminal. | Subcommand | What it does | When to use | |------------|-------------|-------------| | `search` | Search papers by query with field prefixes and Boolean operators | Finding papers on a topic, exploring recent work | | `get_paper` | Retrieve full metadata for one or more papers by ArXiv ID | Looking up a specific paper, getting full abstract and links | | `download_paper` | Download a paper's PDF to a local directory | Saving papers for offline reading or RAG indexing | #### Query Syntax | Prefix | Field | Example | |--------|-------|---------| | `ti:` | Title | `ti:transformer` | | `au:` | Author | `au:bengio` | | `abs:` | Abstract | `abs:"reinforcement learning"` | | `cat:` | Category | `cat:cs.CL` | | `all:` | All fields | `all:attention` | Boolean operators: `AND`, `OR`, `ANDNOT` (uppercase). Date filter: `submittedDate:[YYYYMMDD0000 TO YYYYMMDD0000]`. ### Skills & Commands | Skill | What it does | |-------|-------------| | `/arxiv:agent-arxiv-search` | Tool reference and query strategy guide for the arxiv-search agent | ## Workflows **Literature review:** Ask Claude to research a topic via the `arxiv-search` Skill — multiple queries with different field prefixes, category and date filters, curated paper list with summaries. **Paper lookup + download:** Use `get_paper` to fetch metadata for known ArXiv IDs, then `download_paper` to save PDFs locally. **Exploratory search:** Start broad with `all:keyword`, then narrow with `ti:` and `cat:` prefixes once you know the relevant categories. ## Troubleshooting <details> <summary>No results returned</summary> ArXiv search is exact-match on field prefixes. Try: - Use `all:keyword` instead of a bare query - Check Boolean operators are uppercase (`AND`, not `and`) - Remove date filters to broaden the search - Simplify the query — fewer conditions return more results </details> <details> <summary>ArXiv API rate limiting</summary> ArXiv recommends at least 3 seconds between requests. If you receive empty responses or connection errors after rapid successive searches, wait a few seconds before retrying. The plugin does not add automatic delays. </details> <details> <summary>PDF download fails or saves to wrong path</summary> - `output_dir` must be an absolute path that exists on your filesystem (e.g., `/tmp/papers`) - If the directory does not exist, create it first: `mkdir -p /tmp/papers` - Check that the ArXiv ID is valid — use `get_paper` first to confirm the paper exists </details> ## License MIT