Unifero-CLI is a compact Python toolkit that brings web-search and documentation crawling into a single, easy to use tool. It focuses on safely extracting technical content and code snippets from result pages or documentation sites. The project provides:

• a modern CLI (main.py),
• a FastAPI wrapper (api.py) for HTTP-based automation and testing, and
• a Python class interface (tools.unifero.UniferoTool) for direct programmatic use.

• Features
• Installation
• Quick examples (CLI and API)
• Inputs & Outputs (examples)
• Edge cases, limitations & behavior
• Error handling and retry policy
• Troubleshooting
• Development & tests
• Project structure

• Search mode (DuckDuckGo) with result content extraction.
• Docs mode: crawl a base documentation URL and gather pages + code blocks.
• Code-aware extraction: preserves <pre>/<code> blocks and returns them as fenced Markdown blocks in the output.
• Multiple interfaces: CLI, HTTP API, and programmatic use.
• Networking robustness: connection retries, timeouts and basic backoff for transient failures.
• Output options: pretty JSON, compact JSON, and writing to a file.

Requirements:

• Python 3.8+ (recommended)
• A virtual environment is strongly recommended

Install and set up:

cd /path/to/unifero-cli
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Tip: on macOS and Linux use source .venv/bin/activate. For zsh this is the same. Pick the .venv interpreter for your editor (VS Code) to avoid "import not found" warnings.

CLI: run a quick search

source .venv/bin/activate
python3 main.py --search "Python FastAPI" --limit 3

CLI: crawl docs and save to file

python3 main.py --docs "https://ai-sdk.dev/docs/ai-sdk-ui/chatbot" --output docs_result.json

Start the API server (development):

source .venv/bin/activate
uvicorn api:app --reload

HTTP example (POST body JSON):

{
  "mode": "docs",
  "url": "https://ai-sdk.dev/docs/ai-sdk-ui/chatbot",
  "limit": 2,
  "include_content": true
}

You can POST this to http://127.0.0.1:8000/process and receive the same structure the CLI prints.

1. Search mode input (CLI):

python3 main.py --search "Next.js routing" --limit 2

Search mode JSON output (truncated):

{
  "mode": "search",
  "query": "Next.js routing",
  "results": [
    {
      "title": "Next.js — Routing",
      "url": "https://nextjs.org/docs/routing",
      "snippet": "...routing basics...",
      "content": "# Page title\nSome intro text\n```js\n// code block captured from the page\n```"
    }
  ]
}

2. Docs mode input (HTTP body):

{
  "mode": "docs",
  "url": "https://ai-sdk.dev/docs/ai-sdk-ui/chatbot",
  "limit": 3,
  "include_content": true
}

Docs mode JSON output (truncated):

{
  "mode": "docs",
  "base_url": "https://ai-sdk.dev/docs/ai-sdk-ui/chatbot",
  "results": [
    {
      "url": "https://ai-sdk.dev/docs/ai-sdk-ui/chatbot",
      "title": "AI SDK UI: Chatbot",
      "content": "# AI SDK UI: Chatbot\nSome description...\n```js\nconst chat = useChat(...);\n```",
      "fetched": true
    },
    {
      "url": "https://ai-sdk.dev/docs/ai-sdk-ui/usage",
      "title": "Usage",
      "content": "...",
      "fetched": true
    }
  ]
}

Notes on output fields:

• results: list of pages (search results or crawled docs pages).
• Each result includes url, title, snippet (search mode), and content when include_content is true. content is a Markdown-ready string with fenced code blocks for extracted code.
• fetched: (docs mode) boolean indicating whether the page content was successfully fetched and parsed. If false, the error field may provide a short message.

You may also see several additional fields in the CLI/API outputs and test artifacts (for example in output.txt). These are emitted by the runner/test harness and the core tool to help clients and debugging tools interpret results:

• favicon (per-result): URL to the site's favicon, when available. Useful for UI lists where a compact site icon is shown.
• og_image (per-result): URL of the Open Graph image (og:image) if discovered in the page metadata.
• base_url (top-level for docs): the exact base URL you requested for docs mode. The tool ensures the requested base URL appears in results even if the crawler doesn't discover it.
• status_code (wrapper): the HTTP-like status code returned by the wrapper (e.g., 200 for success, 400 for invalid input). This is not the target site's HTTP code but the wrapper's response code.
• name and request (wrapper): the test-runner or wrapper may produce a name label for the run and echo the request payload so you can trace which input produced the output.
• response (wrapper): when present, this contains the same structured object that the CLI/API returns (the results array, etc.).
• elapsed (wrapper): number of seconds the operation took. Useful for performance logging.
• attempts (wrapper): how many network/operation attempts were made (useful if retries occured).

Example (truncated from a test runner):

{
  "name": "search_minimal",
  "request": {"mode":"search","query":"Next.js routing"},
  "status_code": 200,
  "response": { "query":"Next.js routing", "results": [ ... ] },
  "elapsed": 1.58,
  "attempts": 1
}

How to interpret these fields:

• The response object is the canonical output your client should consume. Wrapper-level metadata (name, status_code, elapsed, attempts, request) are intended for test harnesses, logging, or UI telemetry.
• Per-result favicon and og_image are optional and may be null when the page doesn't declare them or the fetch/parsing failed.
• When fetched is false for a result you can check error for a short message; wrapper metadata still helps diagnose network timeouts or retry behavior.

1. Single-page docs sites (SPA) and client-rendered content:

• The tool fetches server-side HTML only. If a docs site is heavily client-side rendered (content injected via JavaScript), the tool will likely only see the initial shell and will miss the dynamically rendered content. Use the fetched: false/error signals to detect this.

2. Robots/toS and politeness:

• This tool does not implement robots.txt parsing or aggressive rate-limiting. It's intended for small-scale testing. For production crawling, add robots parsing, proper rate limits, and caching.

3. Rate limits and blocking:

• Repeated automated requests to the same host may trigger rate-limiting or blocking. The tool uses a short retry/backoff for transient HTTP failures, but it's not stealthy: respect the target site's policies.

4. Duplicate or noisy content:

• Some pages (headers, footers, menus) contain repeated content; the tool attempts to focus on main <article> or visible containers but may return noise on poorly structured pages.

5. Redirects and base URL normalization:

• docs mode always includes the exact base_url requested as the first result (even if it wasn't discovered by the internal crawler). Redirects are followed by the HTTP client; results will contain the final fetched URL.

6. Maximum crawl size:

• To avoid runaway crawls, limit is capped (default 5, enforced max 10). If you need larger crawls, modify the code carefully and add rate-limiting.

Overview:

• Network calls use a session with retries for transient errors (connection resets, 5xx responses). The retry policy has a small backoff and a limited number of retries.
• Timeouts are applied to HTTP requests. If a request times out, the page is marked with fetched: false and an error message.

Common error fields returned in docs results (per page):

• fetched: boolean (true when parsing succeeded)
• error: short string describing the failure (network error, timeout, parse failure)

Examples:

• When a page times out:

{
  "url": "https://example.com/slow",
  "fetched": false,
  "error": "timeout after 10s"
}

• When a page is client-rendered and contains little server HTML:

{
  "url": "https://spa.example/docs",
  "fetched": false,
  "error": "no usable content found - page may be client-rendered"
}

How the CLI/API surfaces errors:

• CLI prints a non-zero exit code when the top-level operation fails (for example, missing required arguments, invalid JSON input).
• For per-page failures, the operation still returns a 200 OK with the results list containing fetched:false entries; this allows clients to inspect partial success.

• "import fastapi could not be resolved": make sure you selected the .venv interpreter in your editor and ran pip install -r requirements.txt inside the venv.
• If pytest cannot import local modules, set PYTHONPATH=. before calling pytest (or install the package into the venv).
• If extracted content lacks code blocks you expected, the page is likely client-rendered. Consider using a headless browser approach (not included) or point the tool at a direct source page that serves server-side HTML.

Run unit tests:

source .venv/bin/activate
PYTHONPATH=. pytest -q

Run the API integration script (requires the server to be running):

uvicorn api:app --reload
python3 scripts/test_api.py

unifero-cli/
├── assets/              # small assets (logo.svg)
├── main.py              # CLI entrypoint
├── api.py               # FastAPI wrapper
├── requirements.txt     # dependencies
├── tools/
│   ├── __init__.py
│   └── unifero.py       # core logic
├── tests/
│   └── test_main.py
└── scripts/
    └── test_api.py

Contributions welcome. Please include tests for bug fixes or new features. Keep UniferoTool.process_request contract stable if you rely on it from the CLI or API.

MIT-style (open source). Use respectfully and add tests for changes.

A powerful CLI toolkit for web searches and documentation crawling with enhanced code extraction capabilities.

• Smart Web Search: DuckDuckGo-based search with content extraction from result pages
• Documentation Crawling: Crawl documentation sites and extract structured content
• Code Extraction: Enhanced HTML parsing specifically designed to capture code snippets and technical content
• Multiple Interfaces: Modern CLI, legacy JSON input, REST API, and Python library
• Robust Networking: Built-in retries, timeout handling, and error recovery
• Flexible Output: Pretty JSON, compact JSON, or file output

Requirements:

• Python 3.8+
• Virtual environment recommended

Setup:

# Clone or download the project
cd unifero-cli

# Create and activate virtual environment
python3 -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Activate environment
source .venv/bin/activate

# Quick search
python3 main.py --search "Next.js routing"

# Documentation crawl with code extraction
python3 main.py --docs "https://ai-sdk.dev/docs/ai-sdk-ui/chatbot"

# Save results to file
python3 main.py --search "Python FastAPI" --output results.json

# Show all examples
python3 main.py --examples

The enhanced CLI supports intuitive command-line arguments:

Search mode:

# Basic search
python3 main.py --search "Next.js routing"

# Advanced search with options
python3 main.py --search "React hooks" --limit 5 --snippet-len 200 --content-len 3000

# Compact output
python3 main.py --search "Python FastAPI" --compact

# Save to file
python3 main.py --search "Vue.js components" --output search_results.json

Docs mode:

# Basic docs crawl
python3 main.py --docs "https://ai-sdk.dev/docs/ai-sdk-ui/chatbot"

# Advanced docs with options
python3 main.py --docs "https://nextjs.org/docs" --limit 3 --content-limit 2000

# Docs without content (URLs only)
python3 main.py --docs "https://example.com/docs" --no-content

Help and examples:

# Show help
python3 main.py --help

# Show all examples
python3 main.py --examples

For backward compatibility, JSON input is still supported:

# JSON as argument
python3 main.py '{"mode":"search","query":"Next.js routing","limit":3}'

# JSON via environment variable
export UNIFERO_JSON='{"mode":"docs","url":"https://example.com/docs"}'
python3 main.py

# JSON via pipe
echo '{"mode":"search","query":"test"}' | python3 main.py

Use the UniferoTool class directly from Python code:

from tools.unifero import UniferoTool

tool = UniferoTool()
resp = tool.process_request({
    "mode": "search",
    "query": "Next.js routing",
    "limit": 2
})
print(resp)

The process_request method accepts a dict with these keys:

• mode: search (default) or docs
• query: search query (required for search)
• limit: maximum number of results
• url: base url for docs mode
• include_content: whether to fetch page content for docs

Performs DuckDuckGo search and extracts content from result pages.

Parameters:

• query (required): Search query string
• limit: Maximum number of results (default: 5)
• snippet_len: Maximum snippet length (default: 300)
• content_len: Maximum content length (default: 2000)

Crawls documentation sites and extracts structured content with code blocks.

Parameters:

• url (required): Base documentation URL
• limit: Maximum pages to crawl (default: 5, max: 10)
• include_content: Whether to fetch page content (default: true)
• content_limit: Maximum content length per page (default: 2000)

source .venv/bin/activate
PYTHONPATH=. pytest -q

A comprehensive test suite is available for the FastAPI server:

# Start the API server
uvicorn api:app --reload

# In another terminal, run the test suite
python3 scripts/test_api.py

This project includes configuration for easy deployment to Vercel as a serverless FastAPI application.

1. A Vercel account
2. Vercel CLI installed: npm i -g vercel
3. Your project pushed to a Git repository (GitHub, GitLab, or Bitbucket)

1. Test locally first:

   source .venv/bin/activate
   ./deploy.sh

   This will start a local development server at http://localhost:8000

2. Deploy to Vercel:

   # Login to Vercel (one time setup)
   vercel login
   
   # Deploy from your project directory
   vercel
   
   # For production deployment
   vercel --prod

The following files configure Vercel deployment:

• vercel.json: Main Vercel configuration
• runtime.txt: Specifies Python version (3.11)
• .vercelignore: Files to exclude from deployment
• requirements.txt: Python dependencies

Once deployed, your API will have these endpoints:

• GET /health: Health check endpoint
• POST /process: Main API endpoint for processing requests
• GET /docs: FastAPI auto-generated documentation
• GET /redoc: Alternative API documentation

After deployment, you can use your API like this:

# Health check
curl https://your-app.vercel.app/health

# Search request
curl -X POST https://your-app.vercel.app/process \
  -H "Content-Type: application/json" \
  -d '{"mode":"search","query":"Next.js routing","limit":3}'

# Docs request
curl -X POST https://your-app.vercel.app/process \
  -H "Content-Type: application/json" \
  -d '{"mode":"docs","url":"https://nextjs.org/docs","limit":2}'

If your application needs environment variables, you can set them in the Vercel dashboard or via CLI:

vercel env add VARIABLE_NAME

• Import errors: Make sure all dependencies are in requirements.txt
• Timeout issues: Vercel has a 10-second timeout for serverless functions
• Memory issues: Consider reducing content limits for large documents
• Module not found: Ensure proper Python path structure

Before deploying, always test locally:

# Start local development server
./deploy.sh

# Test endpoints
curl http://localhost:8000/health
curl -X POST http://localhost:8000/process \
  -H "Content-Type: application/json" \
  -d '{"mode":"search","query":"test","limit":1}'

unifero-cli/
├── main.py              # Enhanced CLI interface
├── api.py               # FastAPI server wrapper
├── requirements.txt     # Python dependencies
├── tools/
│   ├── __init__.py     # Package initialization
│   └── unifero.py      # Core extraction logic
├── tests/
│   └── test_main.py    # Unit tests
└── scripts/
    └── test_api.py     # API integration tests

1. Fork the repository
2. Create a feature branch
3. Make your changes with tests
4. Ensure all tests pass: PYTHONPATH=. pytest
5. Submit a pull request

Open source - contributions welcome. Keep changes focused and add tests for new functionality.