Claude Scholar
Claude Scholar is a semi-automated research assistant for Claude Code, Codex CLI, Kimi Code CLI, and OpenCode — routing ideation, literature, experiments, and writing through a traceable question-to-writing pipeline, with Zotero and Obsidian integration.
Last reviewed by the paperbanana team on Jul 13, 2026
Install
git clone https://github.com/Galaxy-Dawn/claude-scholar.git /tmp/claude-scholar
bash /tmp/claude-scholar/scripts/setup.shWhat is Claude Scholar?
Claude Scholar is a semi-automated research assistant for academic research and software development, aimed especially at computer science and AI researchers. It is explicitly not an end-to-end autonomous research system that tries to replace the researcher — its stated core idea is that human decision-making stays at the center while the assistant accelerates the workflow around it: which problem is worth pursuing, which papers matter, which hypotheses to test, which results are convincing, and what should be written, submitted, or abandoned remain human calls.
The project routes work through a traceable path — question, evidence, experiment, analysis, claim, writing — and organizes that path into seven primary workflow stages: research ideation (Zotero-integrated), ML project development, experiment analysis, paper writing, paper self-review, submission and rebuttal, and post-acceptance processing. Each stage is backed by its own skills, agents, and slash commands (for example /research-init, /zotero-review, /analyze-results, /rebuttal, /poster), and a shared research-contract.md defines Evidence Records, claim strength, and Claim Promotion Gates so unverified claims cannot silently flow into the writing stage.
Claude Scholar ships as a multi-branch project: the main branch is the Claude Code workflow, with dedicated codex, kimi, and opencode branches for Codex CLI, Kimi Code CLI, and OpenCode respectively. Optional integrations connect it to Zotero (via Galaxy-Dawn/zotero-mcp) for literature import and reading, and to Obsidian for a vault-first project knowledge base spanning Sources, Knowledge, Experiments, Results, Writing, Daily, and Maps folders. Companion Python packages, pubfig and pubtab, add publication-grade figure and table generation to the writing stack.
Core capabilities
Seven-stage academic research pipeline
Covers the full lifecycle from research ideation through ML project development, experiment analysis, paper writing, paper self-review, submission/rebuttal, and post-acceptance promotion, each backed by dedicated skills, agents, and commands.
Zotero-integrated literature workflow
The research-ideation skill and literature-reviewer agent, plus /research-init, /zotero-review, and /zotero-notes commands, turn a vague topic into gap analysis, themed Zotero collections, and structured paper reading notes.
Evidence-gated claims via a shared research contract
A research-contract.md defines Evidence Records, claim strength, and Claim Promotion Gates, connecting ideation, literature synthesis, results reporting, writing, and rebuttal workflows so weak or project-only evidence stays explicit instead of silently hardening into fact.
Obsidian project knowledge base
A vault-first workflow routes durable knowledge into Sources, Knowledge, Experiments, Results, Results/Reports, Writing, Daily, and Maps, with commands like /kb-init, /kb-sync, /kb-lint, and /kb-literature-review for setup and maintenance.
Nature-leaning writing and citation stack
ml-paper-writing, nature-writing, nature-polishing, nature-response, and nature-data skills draft and polish manuscript sections, reviewer responses, and data-availability statements, alongside citation-verification and writing-anti-ai for reference accuracy and human-sounding prose.
Strict experiment analysis and publication figures
results-analysis and results-report run a blocker-first gate (unit of analysis, primary metric, seeds/folds, provenance) before statistical testing and report generation; publication-chart-skill wraps the pubfig and pubtab Python packages for publication-grade figures and tables.
What you can use it for
Starting a new research topic from scratch
Describe a vague topic and ask for a literature-grounded plan; Claude Scholar clarifies the research question, suggests an initial plan or hypothesis list, and can route findings into Zotero or Obsidian.
Reviewing an existing Zotero collection
Point it at a Zotero collection to get paper grouping by theme, a literature synthesis, gap analysis, and candidate research directions worth pursuing next.
Analyzing finished experiment results
Feed it an experiment folder to get metric comparisons, ablation or error-analysis suggestions, and a decision-oriented summary of what is solid, what is weak, and what to run next.
Drafting a paper section or rebuttal
Ask for a related-work section built from current findings and paper notes, or a rebuttal draft for reviewer comments, and get a structured section with claims mapped to evidence.
Maintaining a durable project knowledge base
Use the Obsidian integration to keep literature notes, experiment logs, and writing drafts in one vault-first knowledge base that persists across a long-running project instead of one-off chat output.
How to get started
- 1
Run the recommended full installation
git clone https://github.com/Galaxy-Dawn/claude-scholar.git /tmp/claude-scholar, then bash /tmp/claude-scholar/scripts/setup.sh. The installer is backup-aware and incremental-update friendly: it preserves your existing ~/.claude/CLAUDE.md and installs the repo-managed version as a CLAUDE.scholar.md sidecar for you to manually merge.
- 2
Or install minimally, selectively, or via the plugin marketplace
Copy only a research-focused skill subset, or run /plugin marketplace add Galaxy-Dawn/claude-scholar followed by /plugin install claude-scholar@claude-scholar. Plugin installs cannot distribute rules automatically — clone the repo and copy rules/*.md into ~/.claude/rules or .claude/rules yourself.
- 3
Wire up Zotero and/or Obsidian (optional)
Install Galaxy-Dawn/zotero-mcp per MCP_SETUP.md for paper import and full-text reading, and/or point Claude Scholar at an Obsidian vault per OBSIDIAN_SETUP.md for the project knowledge-base workflow.
- 4
Describe your task in natural language
Start with one concrete request, such as "Help me start research on [topic]" or "Review my Zotero collection on [topic]," rather than trying to memorize the whole system up front.
How it compares to similar skills
Claude Scholar is a lighter, multi-CLI research assistant rather than a single heavyweight pipeline. Depending on how much structure and how many CLIs you need, another skill in this directory may fit better.
Academic Research Skills
Pick Academic Research Skills if you want the most complete Claude-Code-only writing pipeline, with stricter citation-integrity gates than Claude Scholar ships by default.
Scientific Agent Skills
Pick Scientific Agent Skills if your work leans toward domain-specific scientific tooling rather than Claude Scholar’s general CS/AI research pipeline.
Research Skills
Pick research-skills if you want a smaller, more minimal skill set without Claude Scholar’s multi-branch, multi-CLI, Zotero/Obsidian-integrated structure.