Claude research skills, by task
Claude research skills are installable SKILL.md instruction packs that teach Claude Code, Codex and other agents to gather and cite information: multi-source deep research, academic paper search and writing, live web search, and page scraping. The most-installed is just-scrape (245k installs and 56 GitHub stars).
The most-installed research skill from each publisher, then the top picks for each task: deep research, academic papers, web search, scraping and research MCP servers.
Ranked by Skillselion · adoption from the skills.sh registry & GitHub · one skill per publisher · Updated October 28, 2026.
The top 10 Claude research skills
| # | Skill | Stars | Installs |
|---|---|---|---|
| 1 | just-scrapeScrapeGraphAI · ScrapeGraphAI/just-scrapeTop pick Quickly search, crawl, extract structured JSON, or monitor web pages without writing custom scraping code. | 56 | 245k |
| 2 | firecrawl-scrapefirecrawl · firecrawl/cli Instantly turn any webpage or JavaScript-rendered SPA into clean, LLM-ready markdown without writing custom scrapers. | 622 | 83.7k |
| 3 | byted-web-searchskills.volces.com Give their coding agent real-time web search capabilities for verifying facts, prices, policies, and current events before answering. | 0 | 50.4k |
| 4 | tavily-searchtavily-ai · tavily-ai/skills Let their coding agent perform high-quality web searches that return clean, LLM-optimized results with snippets, scores, and metadata. | 474 | 36.3k |
| 5 | Instantly pull structured data from Instagram, Facebook, and hundreds of other platforms via Apify Actors without writing custom scrapers. | 2.4k | 16.4k |
| 6 | parallel-deep-researchparallel-web · parallel-web/parallel-agent-skills When they need exhaustive, multi-source investigation that goes far beyond standard web search. | 73 | 15.7k |
| 7 | nature-academic-searchYuan1z0825 · Yuan1z0825/nature-skills Let their coding agent retrieve peer-reviewed papers, verify DOIs, build search strings, and export citations directly inside Claude Code. | 39.4k | 15.5k |
| 8 | deep-researchaffaan-m · affaan-m/ECC Generate thorough, multi-source cited research reports on any topic using Firecrawl and Exa MCP tools. | 250k | 11.2k |
| 9 | academic-paperImbad0202 · Imbad0202/academic-research-skills Generate high-quality, independently composed bilingual abstracts in English and Traditional Chinese that meet academic journal standards. | 46.4k | 10.9k |
| 10 | scrapebrightdata · brightdata/skills Turn any public webpage into clean markdown or structured data inside agent workflows. | 257 | 10.9k |
Deep research skills for Claude
Deep research skills plan sub-questions, search many sources, and write a cited report instead of answering from one search. They differ in how they budget the search, how strictly they cite, and when they stop.
Full guide: Deep research skills for Claude Code, compared.
- firecrawl-deep-research (firecrawl, 35.5k installs and 151 GitHub stars) - Generate sourced, multi-perspective research reports from the web instead of raw search results.
- deep-research (199-biotechnologies, 10.2k installs and 1k GitHub stars) - Produce citation-tracked, evidence-backed research reports with structured outputs when facing complex analysis questions.
- deep-research (Imbad0202, 7.7k installs and 46.4k GitHub stars) - Run systematic, reproducible literature searches that produce properly formatted annotated bibliographies with full search documentation.
- deep-research (affaan-m, 6k installs and 250k GitHub stars) - Generate thorough, multi-source cited research reports on any topic using Firecrawl and Exa MCP tools.
- deep-research (claude-office-skills, 5k installs and 445 GitHub stars) - Quickly gather, synthesize, and structure deep information on any topic before making product or strategy decisions.
Academic papers and literature review skills for Claude
Academic skills search peer-reviewed papers and arXiv, run literature reviews, draft and review papers, and check citations. Several suites chain these into one paper pipeline.
Full guide: Academic research skills for Claude Code and Codex.
- firecrawl-research-papers (firecrawl, 33.8k installs and 151 GitHub stars) - Quickly gather, scrape, and synthesize academic papers, whitepapers, and technical reports into a coherent literature review.
- academic-paper-reviewer (Imbad0202, 9.9k installs and 46.4k GitHub stars) - Simulate a full academic peer-review panel on research papers, manuscripts, or technical documents before committing to ideas or implementations.
- academic-researcher (shubhamsaboo, 6.7k installs and 133k GitHub stars) - Runs literature reviews, paper summaries and citation formatting for research-backed writing.
- literature-review (affaan-m, 6.4k installs and 250k GitHub stars) - Rapidly synthesize academic papers, arXiv preprints, and technical reports into structured insights before committing to a new feature or product direction.
- academic-search (claude-office-skills, 6.1k installs and 445 GitHub stars) - Quickly locate, analyze, and synthesize peer-reviewed academic papers when building evidence-based features or exploring new technical domains.
Web search skills for Claude
Web search skills give the agent a live search API (Tavily, Perplexity, Exa, Parallel and others) so it can pull current sources mid-task. Most need an API key from the provider.
- tavily-research (tavily-ai, 22.2k installs and 474 GitHub stars) - Run deep, multi-source web research that returns structured, cited reports for competitive analysis, market sizing, comparisons, or literature reviews.
- byted-web-search (kotot, 19.5k installs) - Query Volcengine Doubao web search API for fresh web results when facts are time-sensitive.
- parallel-web-search (parallel-web, 14.5k installs and 73 GitHub stars) - Get fast, cost-effective web research results with multiple parallel queries in one agent call.
- exa-search (affaan-m, 10.1k installs and 250k GitHub stars) - Get accurate, up-to-date neural search results for web content, code examples, companies, competitors, and people directly inside Claude or Cursor.
- web-search (claude-office-skills, 4.8k installs and 445 GitHub stars) - Formulate precise search queries, evaluate sources, and synthesize web findings into actionable insights.
Scraping and crawling skills for Claude
Scraping skills turn web pages into clean markdown or structured JSON, crawl whole sites, and map every URL on a domain, so research runs on full page text instead of search snippets.
- firecrawl-crawl (firecrawl, 80.8k installs and 622 GitHub stars) - Bulk extract structured content from entire websites or documentation sections using the Firecrawl CLI.
- firecrawl-build-scrape (firecrawl, 61.8k installs and 99 GitHub stars) - Add single-page web scraping with Firecrawl directly into their product code.
- tavily-crawl (tavily-ai, 17.6k installs and 474 GitHub stars) - Crawl entire documentation sites or extract content from many related pages and save them locally as markdown files.
- data-scraper-agent (affaan-m, 10.1k installs and 250k GitHub stars) - Instantly generate a complete, scheduled AI data-collection agent for any public source without writing the scraper from scratch.
- web-scraping (jamditis, 6.1k installs and 386 GitHub stars) - Reliably extract clean content from websites, bypass anti-bot measures, and handle paywalls or social media feeds without manual trial and error.
Research MCP servers for Claude
MCP servers connect the agent to a research source as tools: arXiv and Google Scholar for papers, PubMed for medicine, and search or scraping APIs for the open web. Pair one with a skill that tells the agent how to use it.
- Firecrawl MCP Server (firecrawl, 7.4k GitHub stars) - MCP server for Firecrawl — search, scrape, and interact with the web.
- Exa (exa-labs, 5k GitHub stars) - Fast, intelligent web search and web crawling. New mcp tool: Exa-code is a context tool for coding
- Arxiv Mcp Server (blazickjp, 3.1k GitHub stars) - Search arXiv papers, download full text, semantic search, citation graphs, and alerts via MCP.
- Google Scholar MCP Server (JackKuo666, 396 GitHub stars) - MCP server for searching academic papers and citations via Google Scholar.
- Perplexity MCP (jsonallen, 308 GitHub stars) - MCP server providing web search functionality using Perplexity AI's search API.
How do I install a Claude research skill?
- Pick a skill from the ranked table or a task section and open its listing.
- Run the install command on the listing page (for just-scrape:
npx skills add https://github.com/ScrapeGraphAI/just-scrape --skill just-scrape). - If the skill calls a search or scraping API, add that provider’s API key. For a live source such as arXiv, also add its MCP server.
- Ask a research question in Claude Code or Codex; the agent loads the skill when the task matches its description.
Common questions
What is a Claude research skill?
A Claude research skill is a SKILL.md instruction file, often with scripts, that loads into Claude Code or another agent and tells it how to research: which sources to search, how many to read, how to cite them and when to stop. One install command adds it; the agent uses it when a research task comes up.
Which Claude research skill is most installed?
just-scrape leads with 245k installs and 56 GitHub stars: Quickly search, crawl, extract structured JSON, or monitor web pages without writing custom scraping code. The ranking on this page refreshes from the skills.sh registry and GitHub.
Do these research skills work with Codex?
Yes, most of them. A skill is a plain SKILL.md folder, and Codex reads the same format as Claude Code. Skills that call a search or scraping API need that provider’s API key in either agent.
Skill or MCP server: which do I need for research?
A skill is the method (how to plan, search, cite and stop); an MCP server is the connection (tools that query arXiv, Google Scholar or a search API). Many setups use both: an MCP server for the source and a skill for the workflow.
How do I install a Claude research skill?
Pick one from the ranked table or a topic section, then run its install command. For just-scrape: npx skills add https://github.com/ScrapeGraphAI/just-scrape --skill just-scrape