humanMCP Marketplace

A federated listings board across personal humanMCP servers. No accounts. No algorithms. Just humans and their offers.

Live: https://marketplace.humanmcp.net MCP endpoint: POST https://marketplace.humanmcp.net/mcp

What it does

Every human can run their own humanMCP server — publishing listings, offers, and trades. The marketplace crawls these servers and builds a single searchable index of listings.

Think of it as a town square where every stall is independently owned. The marketplace doesn't host content — it indexes and links back to the origin.

Connect an agent

{
  "mcpServers": {
    "humanmcp-marketplace": {
      "type": "http",
      "url": "https://marketplace.humanmcp.net/mcp"
    }
  }
}

MCP Tools

REST API

GET  /servers              → list all registered servers
GET  /servers/{domain}     → server detail
GET  /search?q={query}     → full-text search across all listings
GET  /search?q=bread&type=trade → filter by listing type
GET  /feed                 → recent listings feed (JSON)
POST /register             → register a humanMCP instance
POST /mcp                  → MCP JSON-RPC 2.0 endpoint

Agent discovery

GET  /.well-known/agent.json  → agent profile card
GET  /openapi.json            → OpenAPI 3.1 spec
GET  /healthz                 → status + server/listing counts

Web UI

GET  /                  → homepage — listings + servers
GET  /q?q={query}       → search results
GET  /s/{domain}        → server page

Keyboard shortcuts: / search, j/k navigate, Enter open, d theme, ? help.

Register your server

Any humanMCP instance can join. No account needed — just your domain:

curl -X POST https://marketplace.humanmcp.net/register \
  -H "Content-Type: application/json" \
  -d '{"domain": "yourname-humanmcp.fly.dev"}'

The marketplace will:

1. Verify it's a real humanMCP (fetches /.well-known/agent.json)
2. Index all listings from /listings/feed.json
3. Add to crawl schedule (every 6 hours)

Requirements: your server must expose /.well-known/agent.json and /listings/feed.json.

How crawling works

Your humanMCP server
    ↓
Marketplace fetches /.well-known/agent.json (identity)
    ↓
Fetches /api/profile (author name, bio, tags)
    ↓
Fetches /listings/feed.json (listings)
    ↓
Indexes in SQLite + FTS5
    ↓
Searchable within minutes, re-crawled every 6 hours

Listings stay on your server. Marketplace keeps a search index only.

Stack

• Go (stdlib + modernc.org/sqlite)
• SQLite + FTS5 for full-text search
• Fly.io single machine, 256MB
• No JS framework — server-rendered HTML with keyboard navigation

Run locally

go build -o marketplace ./cmd/server/
./marketplace --seed kapoost.humanmcp.net
# open http://localhost:8080

Key principles

1. Listings only — marketplace indexes offers, trades, services
2. No content hosting — index + link back, always
3. Attribution — author server visible on every result
4. Opt-in — register voluntarily
5. Federated — anyone can run their own marketplace instance
6. Open protocol — humanMCP REST API is the standard

Creative content (poems, essays, images) belongs on the author's server. A separate discovery service for that is planned.

Configuration

Deploy

fly apps create humanmcp-marketplace
fly volumes create marketplace_data --size 1 --region ams
fly deploy

License

MIT