pi-search-tool

A pi extension that provides web search and content fetch tools using search-headless.

Features

• search — Search the internet using Brave Search (default). Returns titles, URLs, and snippets in markdown format.
• fetch — Fetch and extract clean content from any URL. Uses multiple extraction methods (Obscura, HTTP+Readability, Playwright) for best results.

Prerequisites

This extension requires search-headless to be installed:

cd ~/dev/search-headless
./install.sh

This installs the search-web and fetch-content CLI commands globally.

Installation

Add this extension to your pi configuration:

{
  "extensions": ["pi-search-tool"]
}

Or install via npm:

npm install pi-search-tool

Usage

Search

Search the web for information:

search(query: "best practices for REST API design", limit: 10)

Returns markdown-formatted results with titles, URLs, and snippets.

Fetch

Fetch full content from a URL:

fetch(url: "https://example.com/article", maxChars: 50000)

Returns clean, readable markdown content extracted from the page.

How It Works

1. Search calls search-web --format markdown from search-headless
2. Fetch calls fetch-content --format markdown from search-headless
3. Both tools run synchronously and return results directly
4. Output is in markdown format for easy reading by LLMs

Tool Guidelines

When to Use Search

• Finding relevant web pages for a topic
• Discovering documentation, tutorials, or examples
• Researching best practices or solutions

When to Use Fetch

• Reading full content of URLs found via search
• Accessing primary sources in depth
• Extracting clean content from complex web pages

Configuration

The extension uses sensible defaults:

• Search limit: 10 results (configurable, max 20)
• Fetch max chars: 50,000 characters (configurable, max 200,000)
• Timeout: 30 seconds (max 2 minutes)

License

Apache-2.0