duckduckgo-mcp-server

Executes web searches against DuckDuckGo's HTML interface and returns formatted results specifically optimized for LLM consumption. The implementation queries DuckDuckGo directly (avoiding API keys), parses HTML responses, removes ad content and redirect URLs, and structures results with titles, URLs, and snippets in a format that LLMs can easily consume and reason about. Rate limiting (30 req/min) prevents service abuse while maintaining responsiveness.

Retrieves full webpage content from a given URL and parses HTML into clean, LLM-readable text. The implementation uses HTTP requests to fetch raw HTML, applies HTML parsing and text extraction (removing scripts, styles, navigation elements), and formats the output for optimal LLM consumption. Rate limiting (20 req/min) prevents overwhelming target servers while maintaining throughput for content analysis workflows.

Initializes and manages a FastMCP server instance that exposes search and content-fetching tools to MCP-compatible clients. The implementation uses FastMCP's @mcp.tool() decorator pattern to register callable Python functions as remote tools, handles tool invocation routing, manages server lifecycle (startup/shutdown), and provides error handling and logging. The server identifier 'ddg-search' enables client discovery and tool binding.

Implements independent rate limiting for search (30 req/min) and content-fetching (20 req/min) tools using request throttling. The implementation tracks request timestamps per tool, enforces per-minute quotas, and delays requests that exceed limits to maintain compliance without rejecting calls. Rate limits are applied at the tool invocation layer, ensuring fairness across concurrent LLM clients and preventing service abuse.

Processes DuckDuckGo search results and fetched webpage content to remove advertisements, tracking redirects, and boilerplate elements. The implementation identifies and strips ad content from search results, cleans DuckDuckGo redirect URLs to expose actual target URLs, removes script/style tags and navigation elements from HTML, and formats remaining content for LLM consumption. This ensures LLMs receive clean, actionable information without noise.

Enables the DuckDuckGo MCP server to integrate with Claude Desktop through the Model Context Protocol, allowing Claude to invoke search and content-fetching tools directly. The implementation exposes the FastMCP server over stdio (standard input/output), implements MCP protocol message handling (JSON-RPC), and registers tools in Claude Desktop's configuration. This provides seamless tool access without custom UI or API management.

Provides multiple installation and deployment pathways for the DuckDuckGo MCP server: Smithery (simplified MCP server registry), pip package installation, and Docker containerization. Each deployment method handles dependency management, environment configuration, and server lifecycle differently, enabling developers to choose based on their infrastructure and operational preferences. Deployment options are documented with setup instructions for each method.

Implements error handling across search and content-fetching operations with graceful degradation and informative error messages. The implementation catches network errors, parsing failures, rate-limit violations, and malformed inputs, returning structured error responses that LLMs can interpret and act upon. Result formatting ensures consistent output structure (titles, URLs, snippets for search; cleaned text for content) regardless of input variation.