AnySearch: Give Your AI Agent Real-Time Web Search

Your AI agent is smart, but it cannot see the internet. It writes code from memory, answers questions from training data, and hallucinates when it runs out of context. AnySearch fixes this by giving your agent a real-time search engine — web search, vertical domain queries, batch search, and full-page content extraction, all through a single skill.

What AnySearch Does

AnySearch is a search skill that adds four capabilities to your AI agent:

1. Web search — general-purpose queries, like asking Google from your agent
2. Vertical domain search — structured lookups for stocks (Stock:), CVEs, DOIs, patents, and more
3. Batch search — run multiple queries in parallel in a single call
4. Page extraction — fetch a URL and return its content as Markdown

It works with Claude Code, Cursor, Codex, OpenCode, and any agent that supports the SKILL.md standard. One install, cross-platform.

Installation

AnySearch bundles CLI tools in four runtimes: Python, Node.js, Bash, and PowerShell. You only need one that works on your machine.

# Download
curl -L -o anysearch-skill.zip \
  https://github.com/anysearch-ai/anysearch-skill/archive/refs/heads/main.zip

# Unzip and install
unzip anysearch-skill.zip
mv anysearch-skill ~/.claude/skills/anysearch

Adjust the target directory for your agent:

Detect Your Runtime

Run the doc command with each available runtime to find the one that works:

# Try Python first (recommended)
python3 ~/.claude/skills/anysearch/scripts/anysearch_cli.py doc

# Fallback: Node.js
node ~/.claude/skills/anysearch/scripts/anysearch_cli.js doc

# Fallback: Bash
bash ~/.claude/skills/anysearch/scripts/anysearch_cli.sh doc

The first one that runs without errors is your runtime. Save it to runtime.conf so the agent remembers:

echo "Runtime: Python" > ~/.claude/skills/anysearch/runtime.conf
echo "Command: python3 ~/.claude/skills/anysearch/scripts/anysearch_cli.py" \
  >> ~/.claude/skills/anysearch/runtime.conf

API Key (Optional but Recommended)

AnySearch works without an API key using anonymous access, but rate limits are lower. To get a free key:

1. Visit anysearch.com/console/api-keys
2. Create an account and generate a key
3. Add it to your skill:

cp ~/.claude/skills/anysearch/.env.example ~/.claude/skills/anysearch/.env
# Edit .env and set: ANYSEARCH_API_KEY=your_key_here

Key priority: --api_key CLI flag > .env file > environment variable > anonymous.

Real Workflow: General Search

Your agent needs to look up current information it was not trained on.

Search the web for the latest changes in the Next.js 16 App Router.
Focus on breaking changes from version 15 to 16.

The agent calls AnySearch internally:

python3 ~/.claude/skills/anysearch/scripts/anysearch_cli.py \
  search "Next.js 16 App Router breaking changes" --max_results 5

You get structured JSON results with titles, URLs, and snippets. The agent synthesizes these into a useful answer grounded in real sources.

Real Workflow: Vertical Domain Search

You need structured lookups that general web search handles poorly — stock prices, CVE details, academic papers.

Look up the current stock price and recent earnings for NVDA.

AnySearch first calls list_domains to discover that finance queries use a structured sub-domain, then searches with the correct format. The result is more precise than a generic web search because the query structure matches the domain.

python3 ~/.claude/skills/anysearch/scripts/anysearch_cli.py \
  search "NVDA earnings Q1 2026" --content_types data

Supported vertical domains include finance, academic (DOI lookups), travel (IATA codes), security (CVE), patents, and more.

Real Workflow: Batch Search

You need to research multiple independent topics at once.

Research three topics in parallel:
1. What is Model Context Protocol (MCP)?
2. How does Claude Code handle skills?
3. What are the top Agent Skill collections on GitHub?

Instead of three sequential searches, AnySearch runs them in one call:

python3 ~/.claude/skills/anysearch/scripts/anysearch_cli.py batch_search \
  --queries '[
    {"query":"Model Context Protocol MCP","max_results":3},
    {"query":"Claude Code skills system","max_results":3},
    {"query":"top Agent Skill collections GitHub 2026","max_results":3}
  ]'

This is faster than three separate calls and reduces round-trips.

Real Workflow: Page Extraction

The agent finds a useful URL but needs to read its full content, not just a snippet.

Read the full content of https://docs.anthropic.com/en/docs/claude-code/skills
and summarize the key points.

AnySearch extracts the page and returns Markdown:

python3 ~/.claude/skills/anysearch/scripts/anysearch_cli.py \
  extract "https://docs.anthropic.com/en/docs/claude-code/skills"

The output is already Markdown — no format conversion needed. The agent reads it directly and summarizes.

AnySearch vs. Built-in WebFetch

Most agents have a built-in WebFetch tool. Here is how they compare:

Use WebFetch when you already know the URL. Use AnySearch when you need to find information first, or when you need structured domain-specific results.

Tips

• Start with doc only once. After runtime.conf exists, the agent uses the stored command directly. Running doc on every activation wastes tokens.
• Use --max_results wisely. Five results is usually enough. More results mean more context to process.
• For vertical domains, always call list_domains first. The structured query format produces significantly better results than general web search for domain-specific topics.
• Batch search saves time. If your agent has three independent queries, batch them into one call instead of making three sequential searches.

When Not to Use AnySearch

Skip AnySearch if your queries are entirely about code in the current repository — your agent can read files directly. Also skip it if you are in an air-gapped environment where outbound API calls are not allowed.

Discover more search and research skills on the SkillMap leaderboard.