bahayonghang/alphaxiv-paper-lookup

AlphaXiv Paper Lookup

Fetch structured AI overviews for arXiv papers. No auth required.

Workflow

Step 1: Extract the paper ID

Parse the paper ID from whatever the user provides:

| Input | Paper ID | | ------------------------------------------ | -------------- | | https://arxiv.org/abs/2401.12345 | 2401.12345 | | https://arxiv.org/pdf/2401.12345 | 2401.12345 | | https://alphaxiv.org/overview/2401.12345 | 2401.12345 | | 2401.12345v2 | 2401.12345v2 | | 2401.12345 | 2401.12345 |

Step 2: Resolve the paper

curl -s "https://api.alphaxiv.org/papers/v3/{PAPER_ID}"

On platforms without curl, use WebFetch with the same URL.

Extract versionId from the JSON response. This is the UUID needed for the next call.

If this returns 404, the paper hasn't been indexed on alphaxiv yet.

Step 3: Fetch the AI overview

curl -s "https://api.alphaxiv.org/papers/v3/{VERSION_ID}/overview/{LANG}"

On platforms without curl, use WebFetch with the same URL.

The response contains:

• intermediateReport — the machine-readable report (structured text, best for LLM consumption)
• overview — the full markdown blog post (human-readable)
• summary — structured summary with fields: summary, originalProblem, solution, keyInsights, results
• citations — list of cited papers with titles and justifications

Prefer intermediateReport when available — it's specifically formatted for machine consumption. Fall back to summary fields if intermediateReport is null.

Step 4: If you need more detail, fetch the full text

If the intermediateReport, summary, and overview fields don't contain the specific information the user is asking about (e.g. a particular equation, table, or section), fetch the full paper text:

curl -s "https://alphaxiv.org/abs/{PAPER_ID}.md"

On platforms without curl, use WebFetch with the same URL.

This returns the full extracted text of the paper as markdown. Only use this as a fallback — the overview and intermediate report are usually sufficient.

If this returns 404, the full text hasn't been processed yet. As a last resort, direct the user to the PDF at https://arxiv.org/pdf/{PAPER_ID}.

Output Format

Adapt presentation to user intent:

| User intent | Format | |---|---| | "summarize" / "overview" | 3-5 bullet key contributions, then 1-paragraph summary | | "explain" / "what does it do" | Plain-language explanation, avoid jargon | | "what are the results" | Lead with metrics from results field, then context | | "key ideas" / "main points" | Numbered list of 3-5 insights from keyInsights | | "read this paper" | Full breakdown: problem → method → results → limitations |

Always include paper title and year. Never reproduce the overview blob verbatim — synthesize from intermediateReport or summary fields.

Error Handling

• 404 on Step 2: Paper not indexed. Tell the user it's not available on alphaxiv yet.
• 404 on Step 3: Overview not generated for this paper.
• intermediateReport is null: Use summary and overview fields instead.

Language

Default: en. Auto-detect the user's language and use the matching code:

| Language | Code | |---|---| | English | en | | French | fr | | German | de | | Spanish | es | | Chinese | zh | | Japanese | ja | | Arabic | ar | | Hindi | hi | | Portuguese | pt |

Substitute in Step 3: .../overview/{LANG}. If the user writes in Chinese, use zh automatically — do not ask.