claude-research-pack
Portable Claude research pack for Windows: three install paths (Code+WSL, Code-native, Desktop). Deep-research, paper-capture into Obsidian, lit-status, MCP servers (arxiv/Semantic Scholar/scihub/university-paper-access/obsidian), and v3 cross-device research-continuity skills.
Install / Use
claude mcp add The-Fish-Salmon -- npx -y github:The-Fish-Salmon/claude-research-packIf the server publishes to npm under a different name, use that package instead — check the repo README.
MCP Server
Model Context Protocol server
Quality Score
Category
Education & ResearchSupported Platforms
Our assessment of claude-research-pack
claude-research-pack scores 75/100 on our quality scale, 400th of 433 Education & Research skills we index.
Its MCP Server is 9.1 KB long, well organised into 16 sections with 3 code examples: a thorough specification that gives an agent plenty to work with.
It has 3 GitHub stars, so there is little community track record yet; judge it on its content.
Maintenance, license and trust
- The repository was last updated about 5 months ago. That is recent enough to be usable, but agent tooling moves fast, so check the instructions against your agent's current version.
- Our last check on 2026-08-31 found the source still online.
- No license is declared. By default that means all rights are reserved: you can read it, but reusing or redistributing it is not clearly permitted. Ask the author before building on it commercially.
- Its trust signals score 78/100, with 2 cautions from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
claude-research-pack compared with similar skills
All 4 of these similar skills score higher than claude-research-pack; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| claude-research-pack (this skill)by The-Fish-Salmon | 75 | 3 | 5mo ago | MCP Server |
| Agent-Reachby Panniantong | 100 | 92.4k | 21d ago | CLAUDE.md |
| headroomby headroomlabs-ai | 100 | 74.5k | today | CLAUDE.md |
| last30days-skillby mvanhorn | 100 | 63.6k | 1d ago | CLAUDE.md |
| CowAgentby zhayujie | 100 | 47.2k | today | CLAUDE.md |
Frequently asked questions
- How do I install claude-research-pack?
- Run
claude mcp add The-Fish-Salmon -- npx -y github:The-Fish-Salmon/claude-research-pack. The install tabs above show the steps for each supported agent. - Which AI agents does claude-research-pack work with?
- It is written for Claude Code and Claude Desktop, as a MCP Server file. Other agents that read the same format can often use it too.
- Is claude-research-pack safe to use?
- It declares no license and scores 78/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is claude-research-pack still maintained?
- The repository was last updated about 5 months ago. That is recent enough to be usable, but agent tooling moves fast, so check the instructions against your agent's current version.
Skill content
View source on GitHubClaude Research Pack
A portable bundle that turns any Windows laptop into a literature-aware research workstation: multi-mode deep research, automatic paper download, structured paper capture into an Obsidian vault, and project-aware citation reuse.
Built on top of the academic-research-skills project by Cheng-I Wu (CC-BY-NC 4.0) -- see ATTRIBUTION.md. Reimplemented in Claude-native primitives (skills + hooks + slash commands + MCP servers).
v2 -- three install paths
Pick one based on which Claude product you want to drive:
| | Path A -- Code on WSL | Path B -- Code on Windows native | Path C -- Claude Desktop | |---|---|---|---| | Audience | Linux-comfortable; existing rig | CLI users without WSL | Non-developer / GUI users | | Slash commands | ✅ | ✅ | ❌ (free-text invocation) | | Sub-agent integrity gates | ✅ | ✅ | ❌ -- single-pass + self-critique | | Hooks (handoff, todo persistence, statusline) | ✅ | ✅ | ❌ | | Iron Rules + citation discipline | ✅ | ✅ | ✅ | | Paper capture into vault | ✅ | ✅ | ✅ | | Cross-device research continuity (snapshot/resume via synced vault) | ❌ | ❌ | ✅ -- v3 | | Local-PDF ingestion | ❌ | ✅ -- v5 | ✅ -- v4 | | Interactive research co-pilot loop | ❌ | ✅ -- v5 | ✅ -- v4 | | Mandatory scope confirmation + citation pre-flight | ❌ | ✅ -- v5 | ✅ -- v4 | | One-command install (winget + auto-detect + self-test) | ❌ | ✅ -- v5 | ❌ | | Setup time | 30-45 min | ~5 min (after Claude Code) | 20-30 min |
Decision tree and capability matrix in INSTALL_WINDOWS.md.
Path C honest disclosure: Claude Desktop has no Agent tool, so deep-research runs single-pass with explicit in-context self-critique instead of independent parallel sub-agents. Iron Rules still apply (no fabricated citations) but the integrity gate is mechanically weaker. Read desktop-skills/academic-deep-research/references/desktop_limitations.md before relying on Path C output for publication-grade work.
What's in the box
Skills (full versions -- used by Path A and Path B)
deep-research-- 7-mode research pipeline (full / quick / lit-review / fact-check / socratic / systematic-review / review). Citation-disciplined: every claim must resolve through a real source. Multi-agent: spawns parallel investigators, synthesizer, devil's-advocate. See skills/deep-research/SKILL.md.paper-capture-- Resolve DOI/arXiv/URL/title -> metadata + PDF + structured Obsidian note. Idempotent and de-duped.lit-status-- Read-only library inspector: counts, tags, gaps, citation links to active projects.handoff-- Capture a session snapshot for the next chat.
Skills (lite versions -- used by Path C)
desktop-skills/academic-deep-research-- Same 7 modes, same Iron Rules, same templates, but workflow is serial in one context with explicit=== DEVIL'S ADVOCATE CHECKPOINT N ===self-critique banners. The skill is namedacademic-deep-research(notdeep-research) to avoid colliding with Claude Desktop's built-in deep-research feature. See desktop-skills/academic-deep-research/SKILL.md.desktop-skills/paper-captureanddesktop-skills/lit-status-- Identical to the Code variants (no Agent tool dependency).desktop-skills/capture-research-state,resume-research-state,sync-check,paper-map-- v3 cross-device continuity. Snapshot a Desktop session into a synced Obsidian folder; resume it on another device. See INSTALL_WINDOWS.md for setup. Backed by a stdlib Python helper at tools/research_sync_agent.py bundled into each skill'sbin/at zip time.desktop-skills/ingest-pdf-- v4. Drop a local PDF (or a folder of PDFs); the skill extracts the DOI / arXiv id / title from the file, hands off topaper-capturefor resolution, and moves the original into80_Attachments/papers/{citekey}.pdf. Closes the gap where users had a paper on disk but no DOI handy.desktop-skills/research-copilot-- v4. The interactive loop layer: Orient -> Question -> Suggest -> Synthesize -> Escalate. One conversational move per turn, never invents citations, defers heavy lifting to the other skills. Turns the pack from a one-shot lit-review tool into an ongoing research conversation.
Slash commands (Path A / B only)
/research [--mode <name>] <topic>/capture-paper <doi|arxiv|url|title> [--citekey <key>] [--project <slug>]/lit-map [summary|unread|tags|gaps|orphans|citation-map <citekey>]/status,/port-to-vault
MCP servers (installed by the bundled installer)
arxiv,semantic-scholar,paper-search,paper-mcp-- paper search & metadatauniversity-paper-access-- institutional full-text via Unpaywallscihub-- last-resort PDFobsidian-- read/search/write the Obsidian vault
Two installers ship in this pack:
mcp-servers/install-mcp-servers.sh-- bash, for WSL (Path A)mcp-servers/install-mcp-servers.ps1-- PowerShell, for Windows native and Desktop (Path B and C; pick target with-Target Native|Desktop)
Hooks (Path A / B only -- Desktop has no hooks API)
precompact-handoff.py-- writehandoff_latest.mdbefore context compactionsession-start-context.py-- inject latest handoff + review summary on resumestop-persist-todos.py-- keep TodoWrite items alive across sessionsstatusline.sh(WSL) /statusline.ps1(Windows native) -- bottom-bar statuspaper-mention-detect.py-- off by default -- suggest/capture-paperwhen a DOI is mentioned
Vault templates (all three paths)
PARA-style scaffolding for Obsidian. Drop into a fresh vault to get
30_Literature/, 70_Templates/literature.md, lit-review and research-question
templates.
Install + use
- INSTALL_WINDOWS.md -- full decision tree and walkthroughs.
- USAGE.md -- how to actually use the pack after install: trigger prompts for deep research, auto-download papers, fine-tune search, ingest local PDFs, force quote-backed answers.
The router is:
PS> .\scripts\setup.ps1 -Mode WSL # Path A (default)
PS> .\scripts\setup.ps1 -Mode Native # Path B
PS> .\scripts\setup.ps1 -Mode Desktop # Path C
Quickstart (after install)
Path A / B -- slash commands in Claude Code
/research --mode lit-review "ion-gated transistors for reservoir computing"
/capture-paper 10.1038/s41586-021-03819-2
/lit-map gaps
Path C -- free-text in Claude Desktop
> Do a lit review on ion-gated transistors for reservoir computing.
> Save this paper: 10.1038/s41586-021-03819-2
> What are the gaps in my reading on transistors?
Design principles
- Citation discipline first. A "fact" without a citation that resolves through Semantic Scholar / arXiv / OpenAlex is a hallucination, not a fact. Deep-research is structured to make hallucination expensive -- the Devil's Advocate has block authority on Path A/B and uses an explicit mechanical checklist on Path C.
- Vault is the canonical brain. Drafts land in
00_Inbox/for user curation; onlypaper-capturewrites directly to30_Literature/. Project notes accumulate citekeys; the same papers come back as context on the next project. - One pack, three runtimes. WSL, Windows native, and Desktop ship in the same
pack. Recipients pick by path; nothing is renamed across versions, so v1 users
can
git pullto v2 with zero migration. - Bring your own institutional access.
university-paper-accessuses Unpaywall + your campus IP. If you don't have institutional access, the pack still works via arXiv + (optionally) Sci-Hub.
Versions and provenance
- v5.0 -- Path B (Windows-native Code) overhaul: one-command install
with winget pre-flight, auto-detect vault path + REST API key, automatic
vault bootstrap + Obsidian config, post-install self-test. Plus full
feature parity with Path C:
ingest-pdf,research-copilot, mandatory scoping, citation pre-flight,lit-status next-action. New slash commands/ingest-pdfand/copilot. - v4.0 -- adds local-PDF ingestion (
ingest-pdf), interactive co-pilot (research-copilot), mandatory scope confirmation + post-composition citation pre-flight inacademic-deep-research, and anext-actionmode inlit-status. (Path C only at v4; ported to Path B in v5.) - v3.0 -- cross-device research continuity for Path C
(
capture-research-state,resume-research-state,sync-check,paper-map+tools/research_sync_agent.py). - v2.0 -- adds Path B (Windows native) and Path C (Claude Desktop) on top of the original Path A WSL build.
- v1.0 -- Path A only (Claude Code on WSL).
Built against the upstream Imbad0202/academic-research-skills v3.6.5 (April 2026).
License
CC-BY-NC 4.0. See LICENSE and ATTRIBUTION.md.
Related Skills
Agent-Reach
92.4kGive your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
headroom
74.5kCompress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
last30days-skill
63.6kAI agent skill that researches any topic across Reddit, X, YouTube, HN, Polymarket, and the web - then synthesizes a grounded summary
CowAgent
47.2kOpen-source personal AI assistant & Agent Harness. Plans tasks, runs tools and skills, self-evolves with memory and knowledge. Multi-agent, multi-model, multi-channel. Lightweight, extensible, one-line install.
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
