AgentShare MCP Registry is a canonical, curated directory of Model Context Protocol (MCP) servers designed for autonomous AI agents and developers. It features machine-readable discovery via agent.json, owner analytics, agent-native submission via x402 micropayments, and content-aware tag suggestions.
This post explains why we built it, how agents discover and submit listings, and what’s coming next for registry traffic intelligence.
If you ship an MCP server today, discovery is fragmented:
Agents need machine-readable discovery (agent.json, JSON APIs) and human-readable trust signals (reviewed listings, clear metadata).
We built the registry at agentshare.dev/registry to solve both — without becoming a spammy “list everything” directory.
Every submission starts as pending_review
. We verify GitHub, website, and MCP endpoint URLs before publish.
Approved MCPs get a canonical page:
https://agentshare.dev/registry/{id}
Each page includes share actions (X, LinkedIn, Facebook, copy link, GitHub badge snippet) so builders help distribute traffic organically.
Listing owners can track opens over time, referrers, client types, and countries:
GET /api/v1/mcp/stats/{mcp_id}?token={submission_token}
Agents can pay a tiny USDC micropayment to mint a scoped registry:submit
key, then submit metadata:
POST https://agentshare.dev/api/v1/registry/agent-key
POST https://agentshare.dev/api/v1/registry/submit
X-API-Key: agshp_...scoped...
Human review still applies — payment covers abuse resistance, not instant approval.
We don’t inject random DeFi tags on a clothing MCP. Tags are derived from your name, description, URLs, and optional GitHub repo topics:
POST https://agentshare.dev/api/v1/registry/suggest-tags
Content-Type: application/json
{
"name": "Fashion Inventory MCP",
"description": "Track apparel SKUs and size variants for retail merchandising bots.",
"category": "integrations",
"github_url": "https://github.com/you/fashion-mcp"
}
Platform core tags (mcp
, ai
, agent
) are always suggested; domain tags come from your content.
Autonomous agents should not scrape HTML. Start here:
curl -s https://agentshare.dev/agent.json | jq '.mcp_listings'
Key fields:
| Field | Purpose |
|---|---|
entries_json |
|
| Full verified list | |
summary_json |
|
| Compact summary for scripts | |
detail_url_pattern |
|
| Public profile URL pattern | |
agent_submit_key_url |
|
| x402 scoped key mint | |
suggest_tags_url |
|
| GEO tag suggestions | |
listing_template |
|
| Optimal metadata structure |
Also see the companion repo: github.com/anhmtk/agentshare-mcp
ip_hash
(SHA-256), not raw IPsregistry:submit
keys cannot access DeFi/commerce data APIsIs listing free?
Yes for humans and agents. x402 on agent-key mint is a small anti-spam layer.
How long does review take?
Typically 24–48 hours. Save your status_url
after submit.
Can agents list without a human account?
Yes — mint a scoped key via x402, submit via API, poll status with the submission token.
How is this different from a generic MCP directory?
Curated review, agent-native JSON discovery, owner analytics, and GEO tooling — not an unmoderated dump.
What analytics do owners get?
Views/opens over time, top referrers, user-agent labels, country breakdown (aggregated).
Free listings grow the network. Aggregate traffic intelligence is the long-term layer:
We’re capturing cross-listing traffic internally first (admin aggregate dashboard + TXT/PDF export) before publishing any public datasets.
POST /api/v1/registry/suggest-tags
If you’re building MCP servers for agents, we’d love to see your listing. The agentic web needs directories that speak both human and machine — we’re shipping one step at a time.
Built by AgentShare — Solana DeFi intelligence + curated MCP registry.
Maintainer: @anhmtk.