I Built a Marketplace for AI Agent Skills — ai, marketplace, nextjs, typescript, Here's the Full Stack Breakdown

Caio Rossi — Fri, 03 Apr 2026 16:45:19 +0000

## Why I Built SkillDepot

Every AI developer has the same problem: you build a great skill for your agent — maybe a data analysis pipeline, a code reviewer, or a content generator — and then it sits in your repo, unused by anyone else.

Meanwhile, another developer 3,000 miles away is building the exact same thing from scratch.

I built SkillDepot to fix this. It's an open marketplace where developers publish, discover, and sell AI agent skills.

## The Core Idea: Skills as .md Files

Here's the unconventional choice: skills on SkillDepot are `.md` files. Not Docker containers. Not compiled binaries. Not Python packages.

Why? Because AI agent skills are fundamentally **text instructions**. They're prompts, rules, and workflows that get injected into an agent's context. Making them simple markdown files means:

- **Zero deployment friction** — publish in seconds
- **Framework agnostic** — works with LangChain, CrewAI, AutoGen, Claude SDK
- **Human readable** — buyers can review exactly what they're getting
- **Version controlled** — standard git workflows apply

## The Stack

- **Frontend:** Next.js 16 (App Router), React 19, TypeScript, Tailwind CSS 4
- **Backend:** Supabase (PostgreSQL + Auth + Real-time + RLS on 16 tables)
- **Payments:** Stripe Connect (Checkout + payouts)
- **Hosting:** Vercel free tier
- **Search:** Full-text search with filters (category, framework, rating, trending)

Total monthly cost: ~$1 (just the domain).

## The Economics

This is where it gets interesting. Most marketplaces take 30-40% (Apple, Google, OpenAI GPT Store). SkillDepot takes 10%.

**Creators keep 90% of every sale.**

Why? Because the content is lightweight (.md files), infrastructure costs are near-zero, and the real value is in the network effect — more creators, more skills, more buyers, more creators.

## MCP Integration with Claude

We recently shipped an MCP Connector and submitted it to Anthropic's Connectors Directory. This means SkillDepot skills can be discovered and installed directly from Claude.

The connector exposes 8 tools: search, browse, install, rate, and more — all through OAuth 2.0 with PKCE.

If approved, every Claude user gets access to the full SkillDepot catalog without leaving their workflow.

## What's Next

- **Python SDK** — `pip install skilldepot && skilldepot publish my-skill.md`
- **Visual Builder** — create skills without writing markdown
- **Enterprise tier** — team management, private skills, SSO

We're onboarding the first 50 developers for early access to these features.

## Try It

The platform is live at [skilldepot.dev](https://skilldepot.dev). Browse 3,800+ skills across 20 categories, or publish your own in under 10 minutes.

If you're building AI agents and want to monetize your expertise — or just find production-ready skills faster — check it out. Creators keep 90%.

Questions? Happy to discuss the technical decisions in the comments.

Building the App Store for AI Agent Skills

Caio Rossi — Thu, 02 Apr 2026 20:35:36 +0000

The AI agent ecosystem is fragmenting. LangChain, CrewAI, AutoGen, Claude SDK—each one implements agents differently. And each one forces you to rewrite your tools (skills) in their own format.

You write a web scraper for LangChain. It's great. Then your team wants to use it in a CrewAI agent. You rewrite it. Three months later, someone asks for the same scraper in AutoGen. You rewrite it again.

This inefficiency sparked an idea: what if there was an open marketplace for agent skills that worked across frameworks?

That's SkillDepot. It's not a framework. It's a distribution layer for reusable AI capabilities.

The Problem: Framework Fragmentation

Let's be concrete. Here's a web scraper tool in LangChain:

from langchain.tools import tool
from bs4 import BeautifulSoup
import requests

@tool
def web_scraper(url: str, selector: str) -> list:
    """Scrape HTML and extract data."""
    response = requests.get(url)
    soup = BeautifulSoup(response.content, 'html.parser')
    return [el.text for el in soup.select(selector)]

Same tool in CrewAI:

from crewai_tools import tool
from bs4 import BeautifulSoup
import requests

@tool
def web_scraper(url: str, selector: str) -> str:
    """Scrape HTML and extract data."""
    response = requests.get(url)
    soup = BeautifulSoup(response.content, 'html.parser')
    return str([el.text for el in soup.select(selector)])

The logic is identical. The decorators are different. The return types vary. You can't share the skill across frameworks without refactoring.

Multiply this across hundreds of developers, thousands of skills, and you have a problem: the AI agent ecosystem is reinventing the wheel constantly.

The Solution: Framework-Agnostic Skills

We built SkillDepot as a format + marketplace that removes this friction.

A skill in SkillDepot is a simple injectable .md file. Markdown + code blocks. Framework-agnostic by design.

Here's the same web scraper as a SkillDepot skill:

---
name: web_scraper
version: 1.0.0
framework: langchain, crewai, autogen
author: your_name
description: "Scrape HTML and extract structured data using CSS selectors"
parameters:
  - name: url
    type: string
    description: "Target URL to scrape"
  - name: selector
    type: string
    description: "CSS selector for data extraction"
---

# Web Scraper Skill

Extracts text content from HTML elements matching a CSS selector.

## Installation

pip install beautifulsoup4 requests

## Code

from bs4 import BeautifulSoup
import requests

def web_scraper(url: str, selector: str) -> list:
    response = requests.get(url)
    soup = BeautifulSoup(response.content, 'html.parser')
    return [el.text for el in soup.select(selector)]

That's it. One skill. Works everywhere.

How SkillDepot Works

Step 1: Create a skill

Write your skill as a .md file following the template above. Include metadata (name, framework, author), description, dependencies, code block, and usage example.

Step 2: Upload to SkillDepot

skilldepot upload my_skill.md

Or upload via the web dashboard.

Step 3: Developers discover and use your skill

They find your skill in the SkillDepot marketplace (categorized by type, framework, popularity). They import it:

from skilldepot import get_skill

scraper = get_skill("web_scraper")

The SDK handles versioning, dependency injection, and framework compatibility.

Step 4: You earn money

If you set a price, you get paid. SkillDepot takes 10%, you take 90%.

The SkillDepot SDK

Using the SDK is straightforward:

from skilldepot import Client

client = Client(api_key="your-api-key")

# Get a specific skill
skill = client.get_skill("web_scraper", version="1.0.0")

# Search for skills
results = client.search("data extraction", category="web-scraping")

# List your published skills
my_skills = client.my_skills()

# Publish a new skill
client.publish_skill("./my_skill.md", price=9.99)

Installation:

pip install skilldepot

Why This Matters

For developers: You spend less time rewriting tools. You focus on your agents, not on tool packaging. You discover skills built by others instead of rebuilding from scratch.

For creators: You can monetize your expertise. Write a skill once, sell it to developers across all agent frameworks. 90/10 split—you keep 90%.

For the ecosystem: Skills become a tradable asset. The market discovers which tools are actually useful. Open-source skills proliferate. Specialized skills can be monetized.

Current State

SkillDepot is live with:

3,800+ indexed skills across 20 categories
Free tier + Pro ($29/month for 100K API calls)
Python SDK (TypeScript coming soon)
MCP Connector in development with Anthropic

What's Next

We're focusing on better discovery (improved search, trending skills, quality ratings), expanding framework support, MCP integration with Anthropic's Model Context Protocol, and community features like skill ratings and creator profiles.

Limitations (Being Honest)

The .md format is simple. It works for 80% of skills. Complex multi-file skills require workarounds.
Quality is uneven. We're experimenting with community voting and verification.
Monetization is unproven. The market will tell us if developers want to pay for skills vs. open-source.

Getting Started

Visit skilldepot.dev
Browse 3,800+ skills or create your own
Use the free tier or upgrade to Pro for more API calls

If you're building agent workflows, SkillDepot is designed to save you time and let you focus on what matters: your agents.

What skills do you find yourself rewriting the most? Drop a comment — I'd love to hear what the community needs.

Forem: Caio Rossi