arxiv-paper-get

arxiv-paper-get

Download an arXiv paper to a local workspace. Fetches both PDF and LaTeX source (.tar.gz) in parallel, extracts the source into a source/ subdirectory, and writes a metadata file + report skeleton.

Detection

Trigger when the user sends any of these:

• An arxiv.org/abs/… or arxiv.org/pdf/… URL
• "download this paper", "fetch the paper", "get the arxiv paper", "pull the paper locally"
• Any message containing an arXiv identifier like 2506.01966 or arxiv:2506.01966
• The user says "save this paper" while a paper URL is in the conversation

Do NOT trigger when the user is only asking a question about a paper (e.g., "What does 2506.01966 say about…") without asking to download it.

First-run setup (MUST do before anything else)

Before running the workflow, check whether the arxiv-paper-get CLI is installed:

arxiv-paper-get --help

If the command is not found, do NOT proceed silently. Tell the user:

arxiv-paper-get is not installed yet. Install it with:

uv tool install git+https://github.com/hao1305430263/arxiv-paper-get

This only needs to be done once — it compiles the Python script into a global command.

After installation, verify with arxiv-paper-get --help and continue.

Workflow

Once the CLI is confirmed available:

1. Extract the arXiv URL

From the user's message, extract the full arXiv URL. If the user only gives an ID (e.g., 2506.01966), construct the URL: https://arxiv.org/abs/2506.01966

2. Run the download

arxiv-paper-get "https://arxiv.org/abs/XXXX.XXXXX"

The default base directory is ./papers (relative to CWD). To override:

arxiv-paper-get "https://arxiv.org/abs/XXXX.XXXXX" "/path/to/custom/dir"

3. Parse the JSON output

The command prints a JSON object to stdout on success. Key fields:

Field	Description
paper_dir	Path to the paper workspace
pdf_path	Path to the downloaded PDF (or null)
source_path	Path to the source tarball (or null)
source_extract_dir	Path to extracted source, typically paper_dir/source/ (or null)
report_path	Path to the report skeleton .md
pdf_ok	true if PDF downloaded successfully
source_ok	true if LaTeX source downloaded successfully
source_extracted	true if source was extracted

4. Report to the user

After the command finishes, summarize:

📥 Downloaded: {title}
📄 PDF: {pdf_path or "not available"}
📦 Source: {source_extract_dir or "not available"}
📝 Report skeleton: {report_path}

If the source tarball was not available (common for older papers), note it:

LaTeX source is not available for this paper — PDF only.

Directory structure created

papers/{Paper Title}/
├── {Paper Title}.pdf          # PDF file
├── source/                    # Extracted LaTeX source (if available)
│   ├── main.tex
│   └── ...
├── metadata.json              # Full paper metadata
└── {Paper Title}_report.md    # Report skeleton for note-taking

The LaTeX source tarball is auto-cleaned after successful extraction. If extraction fails, the tarball is kept for manual inspection.

Gotchas

• TeX source may not be available for all arXiv papers (pre-2000, some publishers). The tool handles this gracefully — PDF is still downloaded.
• Write conflicts: the tool never overwrites an existing report file.
• Windows paths: fully supported; the tool uses pathlib.Path throughout. Path-traversal guards use Path.relative_to() for cross-platform safety.
• Offline: the tool requires network access to arxiv.org and export.arxiv.org.

Example session

User: download https://arxiv.org/abs/2506.01966

Agent: [checks: arxiv-paper-get --help → OK]
Agent: *runs `arxiv-paper-get "https://arxiv.org/abs/2506.01966"`*

📥 Downloaded: Matrix Is All You Need
📄 PDF: papers/Matrix_Is_All_You_Need/Matrix_Is_All_You_Need.pdf
📦 Source: papers/Matrix_Is_All_You_Need/source/
📝 Report skeleton: papers/Matrix_Is_All_You_Need/Matrix_Is_All_You_Need_report.md