winget install --id=OleReinartz.GannetMcp -e gannet-mcp is an MCP server for web searching and fetching webpages. Dual-licensed MIT OR Apache-2.0 (see LICENSE-MIT and LICENSE-APACHE in the repository). This winget package installs the portable x64 Windows build.
A Model Context Protocol (MCP) server implementation in Rust for web searching and webpage fetching.
š Web Search: Search the web using multiple providers
ddgs crate)š Webpage Fetching: Fetch and parse webpages with content extraction
ā” Performance
š Security
gannet-mcp/
āāā Cargo.toml # Project dependencies
āāā src/
ā āāā main.rs # Binary entry point
ā āāā lib.rs # Library root
ā āāā service.rs # `service` subcommand (OS service install/run/status)
ā āāā mcp_config.rs # `mcp-config` subcommand (client JSON snippet)
ā āāā server.rs # MCP server (rmcp 3.x #[tool]/#[tool_router])
ā āāā config.rs # Configuration management
ā āāā error.rs # Error types
ā āāā models/ # Data models
ā ā āāā mod.rs
ā ā āāā search.rs # Search request/response models
ā ā āāā content.rs # Webpage content models
ā ā āāā config.rs # Extended config structures
ā āāā services/ # External service integrations
ā ā āāā mod.rs
ā ā āāā search_service.rs # Search providers (DDG, BrightData, Serper, SearXNG)
ā ā āāā fetch_service.rs # Webpage fetching service
ā āāā handlers/ # MCP protocol handlers (active)
ā āāā mod.rs
ā āāā search_handler.rs # Search tool handler
ā āāā fetch_handler.rs # Fetch tool handler
āāā tests/
ā āāā integration_test.rs # STDIO JSON-RPC integration tests
ā āāā fetch_service_test.rs # Mocked fetch tests (mockito)
ā āāā search_provider_test.rs # Search provider + config tests
āāā .env.example # Environment variables template
āāā mcp.json # IDE MCP registration (searxng)
āāā .cursor/mcp.json # Cursor IDE MCP registration
āāā free_search_providers.md # Comparison of free search providers
āāā search_provider.md # Guide to getting free API keys
āāā README.md # This file
# Clone the repository
git clone https://github.com/reinartz/gannet-mcp.git
cd gannet-mcp
# Build release version
cargo build --release
# The binary will be at target/release/gannet-mcp
cargo install --path .
The server can be configured via:
config crate)MCP__)Create a config.yaml:
server:
use_stdio: true
host: "127.0.0.1"
port: 8080
search:
provider: "duckduckgo"
default_limit: 10
max_limit: 100
timeout_secs: 10
rate_limit:
enabled: true
requests_per_minute: 60
max_concurrent: 10
logging:
level: "info"
format: "pretty"
See .env.example for all available options:
cp .env.example .env
# Edit .env with your configuration
gannet-mcp --help
# Run with default settings
gannet-mcp
# With custom log level
gannet-mcp --log-level debug
# Using config file
gannet-mcp --config config.yaml
# Serve MCP over Streamable HTTP (default binds localhost)
gannet-mcp --http
# Custom endpoint path
gannet-mcp --http --mcp-path /api/mcp
sudo gannet-mcp service install # install + enable + start (systemd/launchd/SCM)
gannet-mcp service status # no root required
sudo gannet-mcp service restart # start | stop | uninstall likewise
install points the OS supervisor at the hidden gannet-mcp service run
entry point, which always serves HTTP (never STDIO).
gannet-mcp mcp-config --client claude --stdio --print # emit JSON snippet
Supported clients: claude | cursor | zed | opencode | continue | copilot;
--stdio (default) or --http; --print (default) or --write.
> Security note: HTTP mode has no authentication in v0.2.0 (bearer-token
> auth is a planned follow-up). Bind localhost (the default) and expose it
> via a reverse proxy with TLS if remote access is needed. Avoid binding
> 0.0.0.0 on an untrusted network.
The server provides two main tools:
web_searchSearch the web for information.
{
"name": "web_search",
"arguments": {
"query": "rust programming language",
"limit": 10,
"language": "en"
}
}
Parameters:
query (required): The search query stringlimit (optional): Maximum results (default: 10, max: 100)offset (optional): Pagination offsetlanguage (optional): Language code (en, es, fr, etc.)safe_search (optional): off, moderate, strictregion (optional): Region code for localized resultsweb_fetchFetch and parse a webpage.
{
"name": "web_fetch",
"arguments": {
"url": "https://example.com",
"extract_links": true,
"extract_images": false
}
}
Parameters:
url (required): URL to fetchinclude_raw_html (optional): Include raw HTML in responseextract_links (optional): Extract links (default: true)extract_images (optional): Extract images (default: false)timeout_secs (optional): Request timeout (default: 30)max_content_size (optional): Max content size in bytes (default: 10MB)No configuration required. Uses the ddgs crate (not HTML scraping).
export MCP__SEARCH__PROVIDER=serper
export MCP__SEARCH__API_KEY=your_api_key
docker run -d -p 8081:8080 searxng/searxngexport MCP__SEARCH__PROVIDER=searxng
export MCP__SEARCH__BASE_URL=http://127.0.0.1:8081
Generic Bright Data SERP API (works with any Bright Data SERP zone).
export MCP__SEARCH__PROVIDER=brightdata
export MCP__SEARCH__API_KEY=your_brightdata_api_key
export MCP__SEARCH__SEARCH_ENGINE_ID=serp_api1
search_provider.md ā Detailed guide for getting free API keys (Serper, Bright Data)free_search_providers.md ā Comparison of all free search providers with free tiersThe codebase follows Rust best practices:
SearchProvider trait)rmcp v3.x macros (#[tool], #[tool_router])# Run all tests
cargo test
# Run with output
cargo test -- --nocapture
# Run specific test
cargo test test_name
# Format code
cargo fmt
# Check formatting
cargo fmt -- --check
# Run clippy
cargo clippy -- -D warnings
The server uses a custom error type ServerError with the following variants:
Network: HTTP/network errorsUrlParse: Invalid URL formatJsonSerialize: JSON serialization errorsConfig: Configuration errorsSearchApi: Search provider API errorsRateLimitExceeded: Rate limit exceededInvalidRequest: Invalid request parametersNotFound: Resource not foundTimeout: Request timeoutMcpProtocol: MCP protocol errorsDual-licensed under MIT OR Apache-2.0 ā see LICENSE-MIT and LICENSE-APACHE files for details.
Built with: