AgentRank

Google for AI agents — live search across 25,000+ scored MCP servers, updated daily

AgentRank

Google PageRank for AI agents. A live, daily-updated index of 25,000+ MCP servers and agent tools — scored by real GitHub signals, not just star counts.

npm version npm downloads GitHub stars License: MIT Last commit


Install in 3 Steps

1. Add to your AI tool:

# Claude Code
claude mcp add agentrank -- npx -y agentrank-mcp-server

2. Ask your AI to find a tool:

"Find me an MCP server for database access"

3. Get ranked, current results — automatically.

That's it. No API key, no config, no prompting. Your AI searches the live index whenever it needs a tool.

Cursor, VS Code, Cline, Claude Desktop, Windsurf →


Why AgentRank?

  • Your AI's knowledge is stale. Training data is months old — it can't know if a tool was abandoned last week or if something better shipped yesterday.
  • Stars lie. A repo with 2,000 stars and no commits in 18 months isn't production-ready. AgentRank weighs freshness, issue health, and dependents — not just popularity.
  • 25,000+ tools, scored daily. The index crawls GitHub nightly. Every recommendation reflects what's happening now.
  • Gets smarter with use. Every query surfaces which tools developers actually reach for, sharpening rankings for everyone.

Demo

> search("postgres mcp server")

1. postgres-mcp          Score: 91  ★ 2.1k  Updated: 2d ago
2. mcp-server-postgres   Score: 84  ★ 891   Updated: 5d ago
3. pg-mcp                Score: 71  ★ 432   Updated: 12d ago
> lookup("github.com/modelcontextprotocol/servers")

{
  "name": "modelcontextprotocol/servers",
  "agentrank_score": 97,
  "stars": 14200,
  "last_commit": "1 day ago",
  "dependents": 1840,
  "issue_health": 0.91
}

Links


Repo Structure

crawler/     GitHub crawler — finds and indexes repos nightly
scorer/      Scoring engine — 5-signal composite score (0-100)
mcp-server/  MCP server package published to npm
workers/     Cloudflare Workers API
site/        Astro frontend (agentrank-ai.com)

How the Score Works

Five signals, weighted by signal quality:

SignalWeightWhat it measures
Freshness25%Days since last commit — stale repos decay fast
Issue health25%Closed / total issues — maintainer responsiveness
Dependents25%Repos that depend on this — real-world adoption
Stars15%Raw popularity signal
Contributors10%Bus factor — solo projects score lower

Scores are recomputed nightly from live GitHub data.


MIT License · Built by @comforteagle

Related Servers