DEV Community

Kiprio
Kiprio

Posted on

I Built a Live Registry of 750+ MCP Servers — Here's What I Found

#mcp #claudeai #developer #api

Model Context Protocol (MCP) has exploded in the past six months. Every week, dozens of new MCP servers appear on GitHub — tools that let Claude and other AI assistants connect to databases, APIs, file systems, and basically anything else you can imagine.

The problem? There's no easy way to discover them.

So I built one: kiprio.com/mcp-registry — a live registry that tracks, categorises, and ranks MCP servers from across GitHub and community lists.

Here's what I found after indexing 750+ servers.

The MCP Ecosystem by Numbers

750+ servers indexed (and growing — the tracker runs daily)

Language breakdown:

  • Python: 32% (241 servers)
  • TypeScript: 31% (236 servers)
  • JavaScript: 9% (66 servers)
  • Go: 8% (59 servers)
  • Rust: 4% (30 servers)

The Python/TypeScript split is almost even. Python dominates for data-heavy integrations (databases, analytics, ML). TypeScript leads for browser-based and web service integrations.

Stars distribution:
The median MCP server has under 50 stars. Most are small personal projects. But the top 10% have thousands — these are the tools that solve real problems and get shared.

Most-starred categories:

  1. Developer tools (Claude Code enhancers, terminal integrations)
  2. Workflow automation (n8n, custom pipeline builders)
  3. Database connectors (PostgreSQL, SQLite, Redis)
  4. API integrations (GitHub, Slack, Notion, Linear)
  5. File system and search tools

What's Actually Useful vs. What's Hype

After looking at 750+ servers, some patterns emerged:

The good: Database MCP servers are genuinely useful. Being able to ask "show me the top 10 products by revenue this month" and have Claude query your actual database is compelling. The GitHub MCP server for code review workflows is also solid.

The redundant: There are at least 40 different "weather" MCP servers, 30+ "web search" wrappers, and countless "read a file" implementations. The long tail is very long.

The interesting: The niche ones. There's an MCP server for interacting with Betfair betting markets, one for Ethereum on-chain data, one for competitive intelligence monitoring. These solve real specific problems.

How the Registry Works

The registry scrapes GitHub daily, processes package.json, pyproject.toml, and mcp.json files to confirm these are actual MCP server implementations (not just repos that mention MCP), then ranks them by:

  • GitHub stars (popularity signal)
  • Recent activity (last commit date)
  • Documentation quality (README completeness)
  • Community adoption (dependents, forks)

You can search the full registry at kiprio.com/mcp-registry or use the API:

# Search for database MCP servers
curl "https://kiprio.com/v1/mcp-registry/search?q=database"

# Get top-rated servers
curl "https://kiprio.com/v1/mcp-registry/top?limit=20"

# Get server details
curl "https://kiprio.com/v1/mcp-registry/server/modelcontextprotocol/servers"
Enter fullscreen mode Exit fullscreen mode

The API is free with rate limits. For bulk access and webhooks when new servers are added, there's a Pro tier.

The Most Interesting Finding

MCP adoption is happening faster than documentation.

Most servers work but have minimal README files. The ecosystem is still in "builders building for builders" mode. If you're writing a new MCP server today, a good README with a working example puts you in the top 20% by documentation quality.

The second interesting finding: TypeScript is winning for production-quality servers. The Python servers tend to be more experimental. The TypeScript ones (especially from the core MCP team and larger contributors) are more complete and battle-tested.

What's Next

The registry is updated daily. Next additions:

  • Changelog feed: subscribe to get notified when a server you're watching releases a new version
  • Compatibility matrix: which servers work with which clients (Claude Desktop, Continue.dev, etc.)
  • Quality scores: automated scoring based on test coverage, documentation, activity

If you've built an MCP server and want it indexed, it should be picked up automatically within 24h. If not, submit it here.

Built with Python, FastAPI, and a lot of GitHub API calls. The full data is available via the API.

— The team at kiprio.com — we build APIs and tools for developers.

Top comments (0)