>_Skillful
Need help with advanced AI agent engineering?Contact FirmAdapt
All Posts

What We Learned Building Skillful.sh: Lessons in AI Tool Discovery

Building an AI tool directory that actually helps people find what they need taught us a lot about how developers search, evaluate, and adopt tools. Here's what we learned.

June 9, 2026Basel Ismail
skillful behind-the-scenes tool-discovery lessons

Search Is Only Half the Problem

When we started building Skillful.sh, we thought the main challenge was search. Help people find the right MCP server or AI tool from a catalog of thousands. And search is important. But we quickly learned that finding a tool is only the first step. The harder problem is helping people evaluate whether a tool is worth using.

A search result that shows you a tool name and a one-line description isn't enough. You need to know: does it work well? Is it maintained? Is it secure? Does it fit my specific use case? Those evaluation questions are what actually determine whether someone adopts a tool or keeps scrolling. That's why we invested heavily in security scoring, detailed metadata, and community signals.

Developers Don't Browse, They Hunt

We expected people to browse categories. Some do, but most arrive with a specific problem. "I need a Postgres MCP server" or "I need an agent that can manage GitHub issues." They want to go from problem to solution as fast as possible, which means search relevance and filtering matter way more than category pages and curated lists.

This shaped how we built the MCP server directory and agent listings. Filters for language, framework, security grade, and popularity let people narrow results quickly. The goal is to get from "I need X" to "here are the 3 best options for X" in under 30 seconds.

Trust Signals Matter More Than Features

Given two MCP servers with similar features, people consistently choose the one with more indicators of quality: more GitHub stars, recent commits, a higher security grade, more downloads. Features are table stakes. Trust is the differentiator. That's why platform stats and community adoption metrics are front and center in our listings.

We also learned that negative signals are as important as positive ones. An MCP server that hasn't been updated in six months, has open security issues, or has poor documentation gets flagged. Helping people avoid bad tools is just as valuable as helping them find good ones.

The Long Tail Is Huge

The top 50 MCP servers get most of the traffic, but the long tail of specialized tools is where the most value lives. A Salesforce MCP server matters to every sales team. An obscure server that connects to a specific industrial control system only matters to a handful of people, but for those people, it's transformative. With 137,000+ tools in the directory, serving the long tail well means every search has to work, not just the popular ones.

What's Next

We're still learning. Every week we look at what people search for, what they can't find, and how they use the tools they discover. The trending page reflects what the community is gravitating toward right now. The skills directory keeps growing as people build more specialized capabilities. And we keep refining the discovery experience based on what actually helps people get from "I need a tool" to "I found the right tool" faster.

If you're building AI tools, we'd love to have them listed. The publisher dashboard makes it straightforward to submit and manage your listings.


Related Reading

Search 137,000+ AI tools on Skillful.sh. Browse MCP servers. See what's trending.