Why this is useful

The hardest part of building with MCP isn't writing the code — it's knowing which servers are worth depending on. There are over 25,000 on GitHub, they vary wildly in quality, and many haven't been touched in months.

AgentRank scores every server daily on five signals: stars, freshness (days since last commit), issue health (close rate), contributor count, and downstream dependents. The composite score runs 0–100. A score above 60 means actively maintained and community-validated. Below 40 means proceed carefully.

With the AgentRank MCP server installed in Claude Code, you can ask Claude to check a server's score before adding it as a dependency, search for the highest-ranked option in a category, or surface what's trending this week — all without leaving your editor.

Step 1 — Install the MCP server

Claude Code reads MCP server configuration from a .mcp.json file. You can place this at the project root (project-level) or in your home directory as ~/.mcp.json (user-level, available in every project).

Add the agentrank entry to your .mcp.json:

{
  "mcpServers": {
    "agentrank": {
      "command": "npx",
      "args": ["-y", "agentrank-mcp-server"]
    }
  }
}

No npm install required. The npx -y flag downloads and runs the latest version automatically on first use. Subsequent runs use the cached version.

Step 2 — Verify it's working

After saving .mcp.json, start or restart Claude Code in that directory. The AgentRank MCP server starts as a background process. To verify it loaded correctly, ask Claude:

List the MCP tools you have available. Do you have an agentrank tool?

You should see three tools listed under agentrank:

All three tools use the public v1 API — no API key, no sign-up required. Rate limit is 100 requests per minute per IP, which is more than enough for interactive use.

Step 3 — Prompts that work

Claude calls the AgentRank tools automatically when your prompt is about tool selection or quality. Here are the prompt patterns that work best.

Finding the best tool for a task

What's the best MCP server for querying a PostgreSQL database?
Use AgentRank to find and compare the top options.

Claude will call search("postgresql database"), get back the top-ranked results, and explain the differences — score, stars, freshness, and what each server does.

Checking a specific tool before adding it

I'm considering adding benborla29/mcp-server-mysql to my project.
Can you look up its AgentRank score and tell me if it's well-maintained?

Claude calls lookup("benborla29/mcp-server-mysql") and returns the score with a plain-English verdict: "score 58.4 — solid, active maintainer, 1,800+ stars".

Comparative selection

I need to add web scraping capability to my agent.
Search AgentRank for the top browser automation MCP servers and recommend one
based on the scores and my use case (headless Chrome, need JavaScript rendering).

This is the highest-value use case: Claude searches for options, compares scores, and makes a recommendation grounded in real data rather than training knowledge that may be months out of date.

Gate-check before adding a dependency

Before I commit to using modelcontextprotocol/servers for filesystem access,
check its AgentRank score. If it's below 50, suggest alternatives.

You can use this as a quality gate in your development workflow — Claude checks the score and only proceeds if it meets a threshold you define.

Step 4 — Use the SDK in your project

For programmatic use — scripts, CI checks, agent pipelines — install the official TypeScript SDK. It's a thin wrapper around the AgentRank REST API with full type coverage.

npm install @agentrank/sdk

Search for tools

import { AgentRank } from '@agentrank/sdk';

const ar = new AgentRank();

// Search for the best browser automation MCP servers
const results = await ar.search('browser automation', { limit: 5 });

for (const tool of results.results) {
  console.log(`#${tool.rank}  ${tool.name}  score=${tool.score.toFixed(1)}`);
}
// #1   microsoft/playwright-mcp    score=71.3
// #2   browserbase/mcp-server-browserbase    score=67.8
// #3   MarkusPfundstein/mcp-server-puppeteer score=61.2

Look up a specific tool

import { AgentRank } from '@agentrank/sdk';

const ar = new AgentRank();

// Check a specific tool before adding it to your project
const tool = await ar.getTool('jlowin/fastmcp');

console.log(`Score: ${tool.score}`);    // Score: 89.4
console.log(`Rank: #${tool.rank}`);    // Rank: #1
// Verdict: highly recommended — active maintainer, 6700+ stars, 89% issue close rate

See what's trending

import { AgentRank } from '@agentrank/sdk';

const ar = new AgentRank();

// See what's trending this week
const movers = await ar.getMovers({ direction: 'up', limit: 5 });

for (const mover of movers.results) {
  const delta = mover.rankDelta > 0 ? `+${mover.rankDelta}` : String(mover.rankDelta);
  console.log(`${delta.padStart(4)} spots  ${mover.name}  score=${mover.score.toFixed(1)}`);
}

Real workflows

Pre-commit MCP quality check

Add a script to your project that checks the AgentRank score of every MCP server in your .mcp.json before you commit. Tools below your threshold fail the check.

import { AgentRank } from '@agentrank/sdk';
import mcpConfig from '../.mcp.json' assert { type: 'json' };

const ar = new AgentRank();
const MIN_SCORE = 50;
let failed = false;

for (const [name, config] of Object.entries(mcpConfig.mcpServers)) {
  // Skip servers that aren't GitHub repos (local paths, etc.)
  const pkg = (config as any).args?.find((a: string) => a.includes('/'));
  if (!pkg) continue;

  try {
    const tool = await ar.getTool(pkg);
    const status = tool.score >= MIN_SCORE ? 'PASS' : 'FAIL';
    console.log(`[${status}] ${name}: score=${tool.score.toFixed(1)} (threshold: ${MIN_SCORE})`);
    if (tool.score < MIN_SCORE) failed = true;
  } catch {
    console.log(`[SKIP] ${name}: not in AgentRank index`);
  }
}

if (failed) process.exit(1);

Agent tool selection at runtime

If you're building an AI agent that dynamically selects MCP servers at runtime, use the SDK to rank candidates and pick the best one:

import { AgentRank } from '@agentrank/sdk';

const ar = new AgentRank();

async function pickBestTool(task: string): Promise<string> {
  const results = await ar.search(task, { limit: 3 });
  if (!results.results.length) throw new Error('No tools found for: ' + task);

  const best = results.results[0];
  console.log(`Selected: ${best.name} (score: ${best.score.toFixed(1)}, rank: #${best.rank})`);
  return best.name;
}

// Use it:
const tool = await pickBestTool('query postgres database');
// Selected: benborla29/mcp-server-mysql (score: 58.4, rank: #18)

Get an API key for richer data

The public API returns rank, score, name, and description. With a free API key you get the full signal breakdown — individual sub-scores for stars, freshness, issue health, contributors, and dependents — plus rank history.

{
  "mcpServers": {
    "agentrank": {
      "command": "npx",
      "args": ["-y", "agentrank-mcp-server"],
      "env": {
        "AGENTRANK_API_KEY": "your_api_key_here"
      }
    }
  }
}

import { AgentRank } from '@agentrank/sdk';

// With API key: gets richer data (score breakdown, rank history)
const ar = new AgentRank({ apiKey: process.env.AGENTRANK_API_KEY });

const tool = await ar.getTool('modelcontextprotocol/python-sdk');
// Returns full signal breakdown: stars score, freshness score, issue health, contributors, dependents

Get a free API key at agentrank-ai.com/docs. The free tier supports 1,000 v2 requests per day, which covers most development workflows.

What's next

• Cursor users — see the Cursor integration guide for the same workflow with Cursor-specific config
• Build your own agent — the full API integration tutorial covers OpenAI function calling and LangChain
• Browse the index — agentrank-ai.com/tools for the full ranked leaderboard
• Submit a tool — agentrank-ai.com/submit to add any GitHub repo to the index