AgentRegistry

Overview

AgentRegistry is a curated marketplace for AI agents that launched September 22, 2026 and crossed 512 Hacker News points on day one. It's the most prominent entry in the newly emergent "agent marketplace" category — a class of platforms that didn't exist commercially until mid-2026 and reached production with paying users by late September.

Unlike the MCP Marketplace, which catalogs MCP servers, or AgentHub Enterprise, which targets B2B procurement, AgentRegistry is designed to be the default listing destination for independent agent developers. The pitch: ship an agent, set a price, and let the registry handle distribution, billing, and revenue splits. Agents can be invoked programmatically through a unified registry API or consumed as marketplace cards by any MCP-compatible client.

The registry takes a 15% cut on standard listings and offers a 5% "premium" tier with priority placement and featured listings. Early adopters include dozens of solo developers and several well-funded agent labs.

Features

• Curated listing gate — new agents are reviewed by the registry team before going live; a manual curation step that differentiates it from open registries
• Model-agnostic backends — agents can target Claude, GPT, Gemini, DeepSeek, Mistral, or local models via Ollama without modification
• Built-in billing and revenue splits — Stripe-based payouts with monthly settlement to authors; the registry handles pricing tiers and refunds
• Agent identity — each listed agent has a persistent identity, versioned history, and reputation score derived from usage signals
• Unified invocation API — any registered client can invoke an agent through a single SDK; the SDK handles billing, authentication, and version routing
• MCP-compatible cards — each agent can be exposed as an MCP server card, allowing MCP-first clients to consume the agent's capabilities as tools
• Author analytics — usage volume, conversion, ratings, and revenue dashboards per agent listing
• Open API — developers can list, query, and invoke agents programmatically through a fully documented REST + MCP interface

Installation

# Install the registry client SDK
pip install agentregistry-client
# or
npm install @agentregistry/client

To list an agent, authors create a agent.yaml manifest and submit it through the registry CLI:

agentregistry submit ./agent.yaml

Quick Start

from agentregistry import Client

client = Client(api_key="reg_live_...")

# Discover agents
agents = client.search(query="code review", top=5)
for agent in agents:
    print(agent.name, agent.revenue_per_month, agent.rating)

# Invoke an agent
result = client.invoke(
    agent_id="agent:pr-review/2026/11",
    input={"file": "auth.py", "task": "review for security issues"},
)
print(result.output)

Agent Manifest

name: "PR Review Agent"
version: "2026.11.0"
author: "your-handle"
backend:
  provider: anthropic
  model: claude-opus-4-7
pricing:
  per_invocation_usd: 0.02
  free_tier: 100/mo
tools:
  - filesystem:read
  - git:diff
tags: [code-review, security, pr]
mcp_expose: true

Advanced Features

• Versioned deployments — each agent version is immutable; clients can pin to a version or float to latest
• Reputation scoring — derived from usage volume, review sentiment, and refund rate
• Cross-agent orchestration — agents can invoke other registered agents as sub-calls through the same unified API
• Enterprise SSO and audit logs — AgentHub Enterprise provides the same backend with SOC 2 reporting

Examples

A production deployment typically wires AgentRegistry into an existing workflow:

# Inside an MCP-compatible orchestration
def review_pull_request(pr_url: str) -> Review:
    registry = Client(api_key=os.environ["REGISTRY_KEY"])
    result = registry.invoke(
        "agent:pr-review/2026/11",
        {"pr_url": pr_url, "focus": "security"},
    )
    return Review.from_dict(result.output)

Pros

• Curated gate keeps the marketplace clean
• Revenue handling is first-class; no side billing plumbing
• Model-agnostic backends make agent portability real
• MCP-compatible cards work across all MCP clients
• Author analytics are comprehensive

Cons

• 15% revenue cut is higher than most app-store equivalents
• Curation gate creates latency for new authors
• Still relatively new — smaller catalog than MCP Marketplace
• No native agent debugging/steering from the registry UI yet

When to Use

• You've built an agent and want to monetize it without building your own billing
• You want a curated catalog (as opposed to open listing)
• You target consumers of agent capabilities, not raw MCP tools
• You need model portability across providers

Resources

• Website: https://agentregistry.dev
• GitHub: https://github.com/agentregistry/client
• Docs: https://agentregistry.dev/docs