4get MCP Server

A MCP server that provides seamless access to the 4get Meta Search engine API for LLM clients via FastMCP.

✨ Features

• 🔍 Multi Search Functions: Web, image, and news search with comprehensive result formatting
• ⚡ Smart Caching: TTL-based response caching with configurable size limits
• 🔄 Retry Logic: Exponential backoff for rate-limited and network errors
• 🏗️ Production Ready: Connection pooling, comprehensive error handling, and validation
• 📊 Rich Responses: Featured answers, related searches, pagination support, and more
• 🧪 Well Tested: Extensive test suite including integration tests with real API, unit tests, and more
• ⚙️ Highly Configurable: 11+ environment variables for fine-tuning
• 🎯 Engine Shorthands: Pick a 4get scraper via the engine parameter without memorizing query strings

📋 Requirements

• Python 3.13+
• uv for dependency management

Quick Start

# Install dependencies
uv sync

# Run the server
uv run -m mcp_4get

# Or use mise
mise run

⚙️ Configuration

The server is highly configurable via environment variables. All settings have sensible defaults for the public https://4get.ca instance.

Core Settings

Variable	Description	Default
FOURGET_BASE_URL	Base URL for the 4get instance	https://4get.ca
FOURGET_PASS	Optional pass token for rate-limited instances	unset
FOURGET_USER_AGENT	Override User-Agent header	mcp-4get/<version>
FOURGET_TIMEOUT	Request timeout in seconds	20.0

Caching & Performance

Variable	Description	Default
FOURGET_CACHE_TTL	Cache lifetime in seconds	600.0
FOURGET_CACHE_MAXSIZE	Maximum cached responses	128
FOURGET_CONNECTION_POOL_MAXSIZE	Max concurrent connections	10
FOURGET_CONNECTION_POOL_MAX_KEEPALIVE	Max persistent connections	5

Retry & Resilience

Variable	Description	Default
FOURGET_MAX_RETRIES	Maximum retry attempts	3
FOURGET_RETRY_BASE_DELAY	Base retry delay in seconds	1.0
FOURGET_RETRY_MAX_DELAY	Maximum retry delay in seconds	60.0

🚀 Running the Server

Local Development

uv run -m mcp_4get

Production Deployment

# With custom configuration
export FOURGET_BASE_URL="https://my-4get-instance.com"
export FOURGET_PASS="my-secret-token"
export FOURGET_CACHE_TTL="300"
export FOURGET_MAX_RETRIES="5"

uv run -m mcp_4get

MCP Server Integration

You can integrate the 4get MCP server with popular IDEs and AI assistants. Here are configuration examples:

Cursor IDE

Add this to your Cursor MCP configuration (~/.cursor/mcp.json):

{
  "mcpServers": {
    "4get": {
      "command": "uvx",
      "args": [
        "mcp_4get@latest"
      ],
      "env": {
        "FOURGET_BASE_URL": "https://4get.ca"
      }
    }
  }
}

OpenAI Codex

Add this to your Codex MCP configuration (~/.codex/config.toml):

[mcp_servers.4get]
command = "uvx"
args = ["mcp_4get@latest"]
env = { FOURGET_BASE_URL = "https://4get.ca" }

Note: Replace /path/to/your/mcp-4get with the actual path to your project directory.

🔧 MCP Tools

The server exposes three powerful search tools with comprehensive response formatting:

fourget_web_search

fourget_web_search(
    query: str,
    page_token: str = None,        # Use 'npt' from previous response
    extended_search: bool = False, # Enable extended search mode
    engine: str = None,             # Pick a scraper from the supported engine list
    extra_params: dict = None      # Language, region, etc.
)

Response includes: web[], answer[], spelling, related[], npt

fourget_image_search

fourget_image_search(
    query: str,
    page_token: str = None,   # Use 'npt' from previous response
    engine: str = None,       # Pick a scraper from the supported engine list
    extra_params: dict = None # Size, color, type filters
)

Response includes: image[], npt

fourget_news_search

fourget_news_search(
    query: str,
    page_token: str = None,   # Use 'npt' from previous response
    engine: str = None,       # Pick a scraper from the supported engine list
    extra_params: dict = None # Date range, source filters
)

Response includes: news[], npt

Engine shorthands

All MCP tools accept an optional engine argument that maps directly to the 4get scraper query parameter. This shorthand overrides any scraper value you may include in extra_params.

Value	Engine
ddg	DuckDuckGo
brave	Brave
mullvad_brave	Mullvad (Brave)
yandex	Yandex
google	Google
google_cse	Google CSE
mullvad_google	Mullvad (Google)
startpage	Startpage
qwant	Qwant
ghostery	Ghostery
yep	Yep
greppr	Greppr
crowdview	Crowdview
mwmbl	Mwmbl
mojeek	Mojeek
baidu	Baidu
coccoc	Coc Coc
solofield	Solofield
marginalia	Marginalia
wiby	wiby
curlie	Curlie

If you need to pass additional 4get query parameters (such as country or language), continue to supply them through extra_params.

📄 Pagination

All tools support pagination via the npt (next page token):

# Get first page
result = await client.web_search("python programming")

# Get next page if available
if result.get('npt'):
    next_page = await client.web_search("ignored", page_token=result['npt'])

🐍 Using the Async Client Directly

You can reuse the bundled async client outside MCP for direct API access:

import asyncio
from mcp_4get.client import FourGetClient
from mcp_4get.config import Config

async def main() -> None:
    client = FourGetClient(Config.from_env())
    data = await client.web_search(
        "model context protocol",
        options={"scraper": "mullvad_brave"},
    )
    for result in data.get("web", []):
        print(result["title"], "->", result["url"])

asyncio.run(main())

This allows you to integrate 4get search capabilities directly into your Python applications without going through the MCP protocol.

🛡️ Error Handling & Resilience

Automatic Retry Logic

• Rate Limiting (429): Exponential backoff with jitter
• Network Errors: Connection failures and timeouts
• Non-retryable: HTTP 404/500 errors fail immediately

Error Types

• FourGetAuthError: Rate limited or invalid authentication
• FourGetAPIError: API returned non-success status
• FourGetTransportError: Network or HTTP protocol errors
• FourGetError: Generic client errors

Configuration Validation

All settings are validated on startup with clear error messages for misconfigurations.

📊 Response Format

Based on the real 4get API, responses include rich metadata:

{
  "status": "ok",
  "web": [
    {
      "title": "Example Result",
      "description": "Result description...",
      "url": "https://example.com",
      "date": 1640995200,
      "type": "web"
    }
  ],
  "answer": [
    {
      "title": "Featured Answer",
      "description": [{"type": "text", "value": "Answer content..."}],
      "url": "https://source.com",
      "table": {"Key": "Value"}
    }
  ],
  "spelling": {
    "type": "no_correction",
    "correction": null
  },
  "related": ["related search", "terms"],
  "npt": "pagination_token_here"
}

Development

This project uses several tools to streamline the development process:

mise

mise is used for managing project-level dependencies and environment variables. mise helps ensure consistent development environments across different machines.

To get started with mise:

1. Install mise by following the instructions on the official website.
2. Run mise install in the project root to set up the development environment.

Environment Variable Overrides: You can override any environment variable by creating a .mise.local.toml file in the project root:

[env]
FOURGET_BASE_URL = "https://your-custom-4get-instance.com"
FOURGET_CACHE_TTL = "300"
# Add any other environment variables you want to override

This file is automatically loaded by mise and allows you to customize your local development environment without modifying the shared configuration files.

UV

UV is used for dependency management and packaging. It provides a clean, version-controlled way to manage project dependencies.

To set up the project with UV:

1. Install UV using mise, or by following the instructions on the official website.
2. Run uv sync to install project dependencies.

MCP Server Integration for local development

Cursor IDE

Add this to your Cursor MCP configuration (~/.cursor/mcp.json):

{
  "mcpServers": {
    "4get": {
      "command": "uv",
      "args": [
        "run",
        "--project",
        "/path/to/your/mcp-4get",
        "-m",
        "src"
      ],
      "env": {
        "FOURGET_BASE_URL": "https://4get.ca"
      }
    }
  }
}

OpenAI Codex

Add this to your Codex MCP configuration (~/.codex/config.toml):

[mcp_servers.4get]
command = "uv"
args = ["run", "--project", "/path/to/your/mcp-4get", "-m", "src"]
env = { FOURGET_BASE_URL = "https://4get.ca" }

Note: Replace /path/to/your/mcp-4get with the actual path to your project directory.

🧪 Testing

Comprehensive test suite with unit, integration, and performance tests:

# Run all tests
uv run pytest

# Run only fast unit tests (exclude integration)
uv run pytest -m "not integration"

# Run integration tests with real 4get API
uv run pytest -m integration

# Run with coverage
uv run pytest --cov=src

# Run specific test categories
uv run pytest tests/test_cache.py      # Cache behavior tests
uv run pytest tests/test_client.py     # Client and retry logic tests
uv run pytest tests/test_integration.py # Real API integration tests

Test Categories

• Unit Tests: Fast, deterministic tests using mock transports
• Integration Tests: Real API tests with rate limiting and resilience validation
• Cache Tests: TTL expiration, eviction policies, concurrent access
• Retry Tests: Exponential backoff, error handling, timeout scenarios
• Configuration Tests: Validation logic and environment variable parsing

The tests follow FastMCP testing guidelines with comprehensive fixtures and proper isolation.

🤝 Contributing

1. Setup: See Development and Quick Start sections
2. Tests: See Testing section
3. Linting: uv run ruff check
4. Format: uv run ruff format

📄 License

GPLv3 License - see LICENSE file for details.