The MCP marketplace chicken-and-egg problem (and how we are solving it)

Every marketplace has the same problem: sellers wont list until there are buyers, and buyers wont come until there are listings. MCP marketplaces have it worse because the buyers are AI agents, not humans.

The problem

I run marketnow.site — a marketplace for MCP (Model Context Protocol) servers. We have 8,845 skills indexed. All of them were auto-imported from GitHub repos.

Zero human-submitted skills.

Why? Because:

1. Nobody knows we exist (1 GitHub star, 51 npm downloads/week)
2. Sellers dont see buyers (no traffic = no reason to list)
3. Buyers dont see sellers (no unique listings = no reason to visit)
4. The MCP ecosystem is new (most devs dont know what MCP is yet)

This is the classic chicken-and-egg problem, but with AI agents as the buyers.

How we are solving it

1. Auto-import (supply side)

We built a massive indexer that scans GitHub for MCP servers (30+ search queries, 8,845 repos found). Every repo gets a 9-layer security audit automatically. No human action needed.

This gives us supply (8,845 skills) without needing sellers. But these arent curated — theyre whatever exists on GitHub.

2. MCP server as discovery channel (demand side)

Our npm package marketnow-mcp lets any AI agent (Claude, Cursor, Cline) search the marketplace directly from their runtime. When an agent needs a tool, it asks MarketNow instead of Googling.

This creates demand without needing human visitors. Agents are the buyers.

3. New: submit_skill tool (viral loop)

We just added a submit_skill tool to the MCP server. Now when an agent discovers MarketNow, it can submit its own skills:

Agent A searches MarketNow for a tool
Agent A finds a useful skill and installs it
Agent A realizes it has its own tools that others could use
Agent A calls submit_skill with its repo URL
MarketNow audits it (9 layers, free)
Agent As skill is now in the marketplace
Agent B discovers Agent As skill
The loop repeats

This is the viral mechanism. Every agent that uses MarketNow can also contribute to it.

4. New: recommend_skills tool

We added a recommend_skills tool that lets agents describe what they want to do and get matching skills:

Agent: I need to scrape a website
MarketNow: Here are 5 web scraping MCP servers. The top one has Sentinel score 8/10.
Agent: Installing it now.

This makes MarketNow useful immediately — no browsing needed.

5. New: verify_trust tool

Agents can verify each others identity using ATC (Agent Trust Card):

Agent A: Who are you?
Agent B: I am agent.example, here is my ATC: ATC-2026-9880252
Agent A: verify_trust(ATC-2026-9880252)
MarketNow: Valid. Sentinel score 10/10. Risk: low.
Agent A: OK, I trust you. Lets interact.

This creates a trust network where MarketNow is the CA (Certificate Authority).

The 8 tools now available

1. search_skills — find skills by keyword
2. get_skill — get full details
3. list_categories — browse categories
4. get_manifest — marketplace metadata
5. get_install_command — get install command (all skills FREE)
6. verify_trust — verify an Agent Trust Card
7. submit_skill — submit your MCP server (FREE, 9-layer audit)
8. recommend_skills — get AI-powered recommendations

What happens next

If 10 agents install marketnow-mcp and each submits 1 skill, we have 10 new human-submitted skills. If 100 agents do it, 100 skills. If 1000 agents do it, we have a real marketplace.

The key insight: we dont need human traffic. We need agent adoption. And agents adopt tools that are useful — MarketNow is useful because it has 8,845 skills + security audits + trust verification.

Try it

npm install marketnow-mcp (v1.5.0)

Add to your agent config:
{
mcpServers: {
marketnow: {
command: npx,
args: [-y, marketnow-mcp]
}
}
}

Then ask your agent: Find me a tool for scraping websites.

Live: https://marketnow.site
GitHub: https://github.com/edgarfloresguerra2011-a11y/marketnow

Edison Flores, AliceLabs LLC