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
engineparameter 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
Caching & Performance
Retry & Resilience
🚀 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.
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 authenticationFourGetAPIError: API returned non-success statusFourGetTransportError: Network or HTTP protocol errorsFourGetError: 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:
- Install mise by following the instructions on the official website.
- Run
mise installin 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:
- Install UV using mise, or by following the instructions on the official website.
- Run
uv syncto 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
- Setup: See Development and Quick Start sections
- Tests: See Testing section
- Linting:
uv run ruff check - Format:
uv run ruff format
📄 License
GPLv3 License - see LICENSE file for details.
