Search1API CLI: Give Your AI Agent a Real Search Tool

Claude Code, Codex, OpenClaw — today's AI agents can all search the web. But if you've actually used them, you know their built-in search is pretty limited.

They return generic web results. They can't search Reddit threads, GitHub repos, arXiv papers, or YouTube videos. Their web fetching handles simple article pages fine, but falls apart on anything more complex.

It's not that the agents aren't smart enough. Their search tools just aren't good enough.

The Problem with Built-in Search

Most agents come with Web Search and Fetch tools that have two key limitations:

Search is one-dimensional. You get generic web results, and that's it. Want your agent to search Reddit for discussions about a framework? Find the latest papers on arXiv? Look up relevant GitHub repos? Built-in tools can't do that.

Fetching is shallow. Basic Fetch grabs simple HTML and calls it a day. But the real web is messier than that — YouTube content lives in transcripts, not page markup. Reddit posts and comment threads need dedicated parsing. X (Twitter) content requires special handling. PDF links need proper extraction. Most Fetch tools just give up.

The result: your agent looks like it can search, but what it finds is often too shallow, too narrow, or just not useful enough.

What We Built Differently

Search1API CLI (short command: s1) is a search tool designed to solve exactly these problems.

13+ Search Engines, One Command

Different questions need different sources. s1 puts 13+ search engines behind a single command, and your agent picks the right one for the job:

# General web search
s1 search "Next.js server actions"

# Search Reddit discussions
s1 search "is Rust worth learning in 2026" -s reddit

# Search GitHub repos
s1 search "vector database" -s github

# Search arXiv papers
s1 search "transformer architecture" -s arxiv

# Search YouTube videos
s1 search "how to deploy on Cloudflare" -s youtube

# Scope to specific sites
s1 search "RAG best practices" --include arxiv.org github.com

Supported engines: Google, Bing, DuckDuckGo, Yahoo, Reddit, GitHub, YouTube, arXiv, X, WeChat, Bilibili, IMDb, and Wikipedia.

Purpose-Built Extraction for Complex Sites

This is where we're fundamentally different from generic Fetch tools. Instead of blindly pulling HTML, we've built dedicated extraction for different types of content:

• YouTube — pulls the actual transcript, not page markup
• Reddit — extracts full posts with comment threads
• X (Twitter) — handles tweet content extraction
• Documents — extracts content from PDFs and other document formats
• Regular pages — intelligently extracts body content, filtering out navbars, ads, and noise

# Get YouTube video content (returns the transcript, not HTML)
s1 crawl https://youtube.com/watch?v=xxx

# Get a Reddit discussion
s1 crawl https://reddit.com/r/rust/comments/xxx

# Search and auto-crawl the top 3 results
s1 search "how to implement RAG" -c 3

What your agent gets back is clean, usable content — not raw HTML that needs another round of processing.

News Aggregation

s1 news "AI regulation" -t day
s1 news "tech layoffs" -s hackernews

Pull the latest news from Google News, Bing News, HackerNews, and more. Filter by recency — last day, month, or year.

Trending Topics

s1 trending github
s1 trending hackernews -n 20

See what's hot in the developer community right now.

Using It with Your Agent

Platforms That Support Skills (Recommended)

More and more agent platforms — Claude Code, Codex, OpenClaw, and others — support the skill protocol. One command to install:

npx skills add fatwang2/search1api-cli

The skill automatically installs the CLI and handles configuration. Once installed, your agent will call s1 whenever it needs to search. Just ask in plain language:

• "Search Reddit for what people think about Bun"
• "What does this YouTube video cover? [link]"
• "Any recent papers on RAG on arXiv?"
• "What's the latest AI news?"

Platforms Without Skill Support

If your agent platform doesn't support skills, install the CLI manually. No Node.js required:

curl -fsSL https://cli.search1api.com/install.sh | bash

Or via npm:

npm install -g search1api-cli

Then log in:

s1 login

Your browser opens automatically, and the API key is saved once you're done. s1 is a standard CLI tool — any agent that can run shell commands can use it.

All commands support --json output for structured results your agent can parse directly:

s1 search "test" --json | jq '.results[0].title'
s1 news "AI" --json

Beyond the CLI

Search1API CLI is the command-line client for the Search1API platform.

On top of the CLI, Search1API also offers a REST API and an MCP Server — the same 13+ search engines, purpose-built content extraction, and news aggregation, through whichever integration fits your use case:

• Skill / CLI — for agents
• MCP — for AI apps that support the MCP protocol
• REST API — for integrating search into your own applications

---

Give your agent better search:

# Platforms with skill support (Claude Code, Codex, OpenClaw, etc.)
npx skills add fatwang2/search1api-cli

# Manual CLI install
curl -fsSL https://cli.search1api.com/install.sh | bash

GitHub: fatwang2/search1api-cli