Paperscope

A Claude Code plugin for academic paper analysis

Paperscope gives Claude the ability to analyze your LaTeX papers. It embeds your paper and its literature into a shared vector space, then measures distances to catch problems that normally require manual reading.

Claude uses these skills automatically when relevant — you don't need to remember commands.

Skills

analyze

when reviewing a paper or checking citations

Citation alignment, novelty detection, argument strength. Checks whether cited papers are actually the closest semantic matches for each claim.

abstract-check

when finalizing for submission

Embeds each section and the abstract, flags sections that are underrepresented. Suggests specific additions.

journal-fit

when choosing where to submit

Fetches recent abstracts from candidate journals via OpenAlex, ranks by semantic similarity to your paper.

revision-diff

when comparing paper versions

Semantic diff between revisions. Measures whether changes moved the paper toward or away from the literature.

when checking bibliography completeness

Searches OpenAlex for semantically similar papers not in your bibliography. Finds missing related work before reviewers do.

cite-check

when verifying references

Verifies DOIs against CrossRef metadata, resolves missing DOIs, catches duplicates and mismatched entries.

Example

You: I've finished revising section 3, can you check the citations?

Claude: Running citation analysis on paper.tex...

Citation alignment: 73% (11/15 contexts aligned)

Line 142: "cortical oscillations maintain coherence"

cites [zhang2019] but best match is [buzsaki2006]

Line 203: Novel claim (max similarity: 0.28)

Your main contribution. Justification in §4 looks solid.

Line 89: Weak paragraph (support: 0.19)

Consider citing [tononi2016] on integrated information.

How it works

1. Text processing — LaTeX is cleaned to plain text and split into ~200-word overlapping chunks.

2. Embedding — Chunks are encoded using sentence-transformers (all-MiniLM-L6-v2, 384 dimensions). Falls back to TF-IDF if the model isn't installed.

3. Similarity — Cosine similarity matrices between your paper and its literature power all 12 analysis tools.

Install

git clone https://github.com/todd866/paperscope.git ~/.paperscope

pip install -r ~/.paperscope/requirements.txt

export PAPERSCOPE_EMAIL="you@university.edu"

# Launch Claude Code with the plugin

claude --plugin-dir ~/.paperscope

Built for this research program. Manages ~1,500 references across 175 .tex files. MIT license. Python 3.8+.