MCP Servers Directory

Model Context Protocol (MCP) is the open standard that connects AI tools - Claude, Cursor, Windsurf, Cline - to external services: databases, APIs, file systems, web search, calendars, project trackers. An MCP server is the bridge: it speaks the protocol on one side, wraps your service on the other.

Before MCP, every AI assistant built its own integration layer. Skills and tools written for Claude didn't transfer to Cursor. MCP changes this: a server written once runs in any MCP-compatible host. Spark indexes 800+ open-source MCP servers - from Anthropic's official connectors to community-built integrations for every major dev tool.

What MCP Servers are on Spark

Official Anthropic-maintained servers cover the most common development needs: filesystem access, Git operations, Brave web search, PostgreSQL, SQLite, Puppeteer, and more. Community-built servers extend coverage to Linear, Notion, Jira, Slack, Stripe, AWS S3, Cloudflare, Supabase, and dozens of specialized data sources. Each entry includes the full installation command, required environment variables, a list of compatible AI hosts, and an activity status reflecting upstream health.

How to install an MCP Server

For Claude Code: add the server entry to .mcp.json in your project root (project-scoped) or to ~/.claude.json (global). Each Spark listing includes a ready-to-paste JSON config block. For Cursor: add the server via MCP settings in the IDE. Most npm and pip packages run in Claude Code, Cursor, and Windsurf without changes.

Choosing the right server

Sort by score - our scoring weights GitHub stars, commit recency, documentation completeness, and community ratings. A score above 70 means the server is actively maintained and widely deployed. The model tag filter narrows results to servers tested with specific Claude versions or connector types (stdio, HTTP, SSE).