tavily-ai/skills
12 skills104k installs4.4k starsGitHub
Install
npx skills add https://github.com/tavily-ai/skillsSkills in this repo
1Tavily SearchTavily Search is an agent skill that runs the Tavily CLI (`tvly search`) so your assistant returns web results optimized for reasoning: snippets, scores, and metadata instead of raw HTML dumps. Solo and indie builders reach for it whenever the task needs current information—competitor pages, regulatory news, library docs, or launch-week announcements—and they do not already have a URL. The skill enforces installing and authenticating the CLI before any command, which keeps sessions reproducible across Claude Code, Cursor, and Codex-style agents. It supports basic and advanced depth, time windows, news topic mode, and domain filtering for focused sweeps. Pair it with downstream Tavily skills for extract, map, crawl, and deep research; search alone is the discovery layer, not full page archival or synthesis.20.8kinstalls2SearchThe Tavily search skill wraps Tavily’s LLM-oriented search API in a small script agents can call with structured JSON. Solo and indie builders use it when they need current web facts, tutorials, news, or domain-specific sources while staying inside Claude Code or similar agents instead of pasting search results manually. Authentication is designed to be low-friction: first run can open browser OAuth via the Tavily MCP auth flow, or you can set TAVILY_API_KEY in settings for API-key mode. Queries support practical filters such as time windows and domain allowlists, which helps validation research, integration docs lookup, and launch-oriented SEO or GEO research. The skill is an integration package, not a standalone MCP server definition in the catalog sense—it ships procedural steps plus search.sh. Confidence is high for placement as a research-oriented agent capability across early journey phases when external grounding is required.12kinstalls3Tavily Researchtavily-research is for solo builders who need defensible answers, not a single SERP snippet. Through the Tavily CLI it gathers sources, analyzes them, and returns a structured report grounded in citations—useful when you are sizing a market, comparing AI coding assistants, or writing a literature-style review before you scope an MVP. The skill is explicit that deep runs take 30 to 120 seconds and that shallow fact checks should use tavily-search instead, which keeps agent loops efficient. It sits late in the Tavily pipeline (search → extract → map → crawl → research), so it rewards chaining rather than one-off chat guesses. Install and login steps are mandatory when `tvly` is absent; the skill forbids silently substituting other research tools. For indie founders, this is the difference between investor-ready competitive notes and hand-wavy positioning copy.11.4kinstalls4Tavily Best PracticesTavily Best Practices is a developer reference skill for production Tavily integrations: choosing between search for live results, extract for URLs, crawl and map for sites, and research for AI-synthesized end-to-end research. Solo builders shipping agentic workflows, autonomous agents, or RAG systems use it while coding so assistants initialize clients correctly, use environment variables for keys, and pick the right method instead of one-size-fits-all search. It pairs installation snippets for tavily-python and @tavily/core with initialization patterns including async clients for parallel queries. The skill does not replace Tavily’s own docs but compresses decision tables and quick-reference calls into agent-invokable procedural knowledge, which reduces bad defaults during Build and speeds up research-heavy Validate or Idea tasks when the same integration is reused.10.9kinstalls5Tavily Extracttavily-extract teaches agents to use the Tavily CLI to grab structured text from specific URLs when the user says extract, read this webpage, or pull the text from a link. It targets JavaScript-rendered sites that resist simple fetches and returns markdown suited for downstream summarization or RAG chunks. Builders can batch many URLs in one invocation or narrow results with query-focused extraction so only authentication or pricing sections surface. The skill enforces a hard prerequisite: install and authenticate `tvly` before substituting other fetch tools, keeping observability and rendering behavior consistent. It sits squarely in discovery and validation work—competitive pages, docs, launch posts—but remains useful during Build when you need to snapshot third-party API references into your repo context.7.9kinstalls6Tavily CliTavily CLI is an agent skill that routes web work through the tvly command-line tool so Claude Code, Cursor, Codex, and similar agents can search, extract page text, crawl documentation, discover URLs on a domain, and run cited deep research. Solo builders use it when native browsing is thin: competitive scans, verifying a link, pulling markdown-friendly content from a URL, or batching doc-site crawls before implementation. The skill documents prerequisites (curl installer or uv/pip), OAuth via tvly login, and JSON responses tuned for downstream reasoning. It is a network integration, not a substitute for git, deploy pipelines, or local refactors. Install once, keep the API key in env, and invoke whenever the user’s ask is explicitly about the public web rather than the repo on disk.7.7kinstalls7Tavily Crawltavily-crawl teaches agents to crawl websites and extract content from many pages using the Tavily CLI (`tvly crawl`), including saving each page as markdown for offline use. Solo builders reach for it when they say crawl the docs, get all pages under /docs, or bulk-extract a vendor site—common during idea research, validating technical feasibility, or stocking build-time reference corpora for RAG and agent context. The skill is strict about tooling: if `tvly` is missing, install via the official script and log in rather than substituting wget or random scrapers. Depth limits, path filters, and semantic instructions help keep crawls bounded on indie budgets. It is a task integration with shell access to Tavily APIs, not a journey-wide methodology; chain it after tavily search/map skills when following the documented Tavily CLI workflow.7.6kinstalls8Tavily Maptavily-map is an agent skill that wraps the Tavily CLI map command so solo builders and coding agents can inventory a website’s URLs without downloading full page content. It fits when you know the site but not the exact subpage—API guides buried in docs, pricing pages on a marketing domain, or blog archives you want to narrow before extraction. The skill mandates installing tvly from cli.tavily.com and logging in rather than substituting other crawlers, and documents JSON output patterns with limits and optional natural-language instructions. For Prism’s journey, it most often appears in Idea research and Validate scope when you are exploring competitors or documentation trees, and again in Build integrations when agents need structured site navigation as part of a Tavily research stack.7.3kinstalls9ResearchThe Tavily research skill gives solo builders an agent-ready path from a plain-language question to a cited synthesis grounded in live web data. You invoke a bundled script with JSON input (topic, optional model tier, and save path), while authentication is handled either by OAuth on first run against the Tavily MCP flow or by setting TAVILY_API_KEY in your agent environment. It fits indie workflows where you need defensible comparisons—competitor landscapes, technology choices, regulatory or market movement—without outsourcing the entire discovery pass to unstructured chat. Because runs can take thirty seconds to two minutes, the skill documentation nudges you toward background execution so your session stays responsive. Use it when narrative quality and source attribution matter more than a one-line answer, and pair it with your own judgment before product or pricing commitments.6.6kinstalls10ExtractExtract is a Tavily-powered agent skill that fetches normalized text from URLs you already trust. Solo and indie builders use it when research, validation, or documentation work needs page content inside the agent loop instead of manual copy-paste or one-off scripts. The skill documents OAuth via the Tavily MCP server on first run (existing Tavily account required) and an API-key alternative in Claude settings. Invocation is JSON-driven: urls array, optional query for topical focus, chunks_per_source, and advanced modes for dynamic sites. It is an integration skill—not a search discovery tool—so you supply links from search, bookmarks, or specs. Prism places it on Idea → research as the canonical shelf, with natural reuse while scoping competitors, reading vendor docs during Build, or archiving launch references. Pair with Tavily search or browse skills when you need to find URLs first, then extract for deep reading.4.9kinstalls11Tavily Dynamic SearchTavily-dynamic-search is Tavily’s agent skill for programmatic web search with strict context isolation. Instead of piping massive raw HTML into the model, you run tvly search and post-process in Python so only deliberate print() summaries return—preserving reasoning quality and token budget. Solo builders reach for it whenever a task needs fresh web facts, competitive snapshots, or filtered excerpts (“find the important parts”) during Idea research, Validate scoping, Launch positioning, or Grow content ideation. The SKILL.md positions it as the default web research skill and documents why naive --include-raw-content blows the window (~8 results × 30–50K chars). It expects local tooling (tvly CLI, Python, optional uv, jq) and fits agentic workflows where the model orchestrates search code rather than reading every navbar and cookie banner.3.8kinstalls12CrawlCrawl is a Tavily-backed agent skill that lets solo builders snapshot web documentation and multi-page sites as markdown on disk instead of copy-pasting pages one at a time. You pass a JSON payload to a shell wrapper—starting with a URL and optionally depth limits, page caps, and output folders—so agents can bulk-ingest vendor docs, competitor help centers, or internal knowledge bases before validation or implementation. Authentication is designed for low friction: first run checks `~/.mcp-auth/` and can open OAuth against an existing Tavily account, with API-key configuration documented for teams that prefer env-based secrets. The skill explicitly targets scenarios where offline access or full-corpus analysis matters, such as comparing integration options during Idea research or freezing third-party API references during Build. It is integration-oriented rather than a generic browser macro: you still choose crawl breadth responsibly and respect site terms. Pair it with analysis or indexing skills after files land locally.3.6kinstalls