paper-wiki

paper-wiki

A personal research-wiki builder for Claude Code. Pipeline-driven paper ingestion, a persistent knowledge-accumulating wiki, and zero LLM API keys — Claude Code itself is the LLM.

paper-wiki turns the firehose of academic publishing into a curated, queryable, compounding wiki — all from inside Claude Code. It fetches papers from arXiv, Semantic Scholar, and (optionally) paperclip's biomedical corpus; scores them against your interests; saves durable notes; and lets Claude synthesize concept articles that link sources together.

It follows a simple three-step model:

① Collect              ② Build knowledge        ③ Use
digest / analyze   →   wiki-ingest          →   wiki-query / wiki-graph
(papers land in        (synthesize concept      (ask your wiki,
 Wiki/papers/)          articles + graph)        cited answers + graph)

Prerequisites

• Python ≥ 3.11 — the plugin self-installs its virtualenv on first session via hooks/ensure-env.sh (uses uv when available, else python3 -m venv).
• Claude Code CLI ≥ 2.1.119 — paper-wiki is a Claude Code plugin; its SKILLs run inside Claude Code, not standalone.
• Optional — Semantic Scholar API key (free): raises the rate limit from ~1 to 100 req/s. Recommended.
• Optional — paperclip for biomedical literature (bioRxiv / medRxiv / PMC). See Biomedical literature.

Install

In Claude Code:

/plugin marketplace add kaneyxx/paper-wiki
/plugin install paper-wiki@paper-wiki

The Python environment self-installs on first session — no manual setup.

Quick start

1. /paper-wiki:setup                    # one-time wizard: vault, topics, API key
2. /paper-wiki:digest                   # pull & score recent papers on your topics
3. /paper-wiki:wiki-ingest arxiv:2401.12345   # fold a paper into concept articles
4. /paper-wiki:wiki-query "what I know about X"   # ask your wiki, get cited answers

The wizard writes ~/.config/paper-wiki/recipes/daily.yaml (your personal recipe) and ~/.config/paper-wiki/secrets.env (API keys, chmod 600). Re-running it is non-destructive (Keep / Reconfigure / Edit one piece).

---

Usage

paper-wiki is a daily loop: collect → build knowledge → use, with periodic maintenance.

every day:   /paper-wiki:digest                  # see what's new
             /paper-wiki:wiki-ingest <id>         # fold 1–3 papers into the wiki
anytime:     /paper-wiki:wiki-query <question>    # use the accumulated knowledge
weekly:      /paper-wiki:wiki-lint                # health-check

Why two steps to "use"? digest only collects papers into Wiki/papers/. wiki-ingest synthesizes them into concept articles. wiki-query / wiki-graph read those concepts — so ingest first, then query.

① Collect papers

/paper-wiki:digest — your main tool. Runs your recipe (topics + keywords) against arXiv + Semantic Scholar, scores and ranks results, writes a dated digest, and drops per-paper notes into Wiki/papers/.

/paper-wiki:digest                 # today, default recipe
/paper-wiki:digest weekly          # a named recipe (recipes/weekly.yaml)
/paper-wiki:digest 2026-05-20      # a specific date

If your recipe sets auto_ingest_top: 3, the top 3 papers are auto-folded into concept stubs at the end of the run — run wiki-ingest on them later to synthesize full articles.

/paper-wiki:analyze — deep-analyze one paper into a structured note. Accepts an arXiv id, URL, s2:<id>, or a fuzzy title.

/paper-wiki:analyze arxiv:2605.05810
/paper-wiki:analyze https://arxiv.org/abs/2605.05810

/paper-wiki:bio-search — search bioRxiv / medRxiv / PMC (requires the paperclip MCP server).

/paper-wiki:bio-search glaucoma vision-language model

② Build knowledge

/paper-wiki:wiki-ingest — the heart of the loop. Reads a source paper and updates (or creates) the concept articles it touches. Folds the new evidence into existing prose without clobbering anything you've marked status: reviewed.

/paper-wiki:wiki-ingest arxiv:2605.06295

/paper-wiki:extract-images — pull real figures from an arXiv source tarball into a paper note (arXiv only).

/paper-wiki:extract-images arxiv:2605.07367

③ Use your wiki

/paper-wiki:wiki-query — ask a natural-language question. Keyword search across concepts and sources; Claude synthesizes an answer grounded only in your wiki, each claim cited with a [[wikilink]].

/paper-wiki:wiki-query what are the failure modes of VLMs on medical images
/paper-wiki:wiki-query which papers use knowledge graphs for reasoning

/paper-wiki:wiki-graph — structured queries over the typed-entity knowledge graph (not free text). Three query types: