---
title: What is the difference between an AI agent registry and an agent marketplace?
url: https://opencontroller.lyzr.com/blog/agent-registry-vs-agent-marketplace
published: 2026-11-17
updated: 2026-11-17
author: Lyzr
---

# What is the difference between an AI agent registry and an agent marketplace?

**An agent registry is the system of record for every agent in your estate: who owns it, which version is running, what it can access, and whether it is approved. An agent marketplace (or internal agent store) is a catalogue where people find and adopt agents. A marketplace is about distribution; a registry is about control. A marketplace without a registry behind it spreads agents faster than anyone can govern them.**

The terms get mixed up because both are lists of agents. The difference is who uses the list and what happens when an entry changes.

## What is an agent marketplace?

A marketplace helps people discover agents and start using them. Public examples exist: Salesforce launched AgentExchange in 2025 for partner-built Agentforce components, and AWS Marketplace added an "AI Agents and Tools" category the same year. Inside enterprises, the equivalent is an internal agent store or gallery in a workspace product, where employees browse approved agents and add them.

A marketplace entry is written for a buyer or user: a description, screenshots, instructions, sometimes pricing. Its success metric is adoption.

## What is an agent registry?

A registry records every agent that exists or runs in your estate, including agents nobody would ever list in a store: background automations, coding agents, vendor agents inside SaaS products, and agents someone built last week without asking. Each entry carries governance data: owner, version history, model, tools and MCP servers, permissions, risk tier, approval state, and runtime status.

A registry entry is written for security, risk and platform teams. Its success metric is coverage: how much of what actually runs is recorded and controlled. Our explainer on [what an AI agent registry is](/blog/what-is-an-ai-agent-registry) and the list of [AI agent inventory fields](/blog/ai-agent-inventory-fields) go into the detail.

## Side-by-side comparison

| Dimension | Agent registry | Agent marketplace |
|---|---|---|
| Purpose | Control and accountability | Discovery and adoption |
| Primary users | Security, risk, platform, audit | Employees, business teams, buyers |
| Coverage | Everything that runs, including unapproved agents | Only agents someone chose to publish |
| How entries arrive | Registration plus automated discovery | Publishing by an author or vendor |
| Key fields | Owner, version, model, tools, permissions, risk tier, status | Description, category, rating, instructions |
| What a change triggers | Re-review, policy check, possible block | Updated listing |
| Lifecycle | Draft to retired, with history kept | Listed or delisted |
| Runtime link | Tied to observability and enforcement | Usually none |

## Why does a marketplace need a registry behind it?

Three reasons.

1. **Approval must come first.** If an agent can be listed before it is risk-rated and approved, the store becomes the fastest way to spread an ungoverned agent. Make "approved in the registry" a precondition for listing.
2. **Versions drift.** A store listing usually shows one agent. The registry tracks which version is actually running, and whether a new prompt, model or tool was added since approval.
3. **Delisting is not stopping.** Removing an agent from a store stops new adoption. It does not stop copies already running. Stopping needs a runtime control linked to the registry entry.

The same pattern applies to tools. Public directories of MCP servers help developers find servers; an internal, governed list decides which ones are allowed. See [how to run a private MCP registry](/blog/run-a-private-mcp-registry).

## How do you connect the two?

1. Make the registry the source of truth; the store reads from it.
2. List only agents whose registry status is approved and whose risk tier allows self-service adoption.
3. Show registry facts in the listing: owner, data accessed, risk tier.
4. When the registry status changes to suspended or retired, delist automatically and trigger the runtime control.
5. Feed adoption data from the store back into the registry, so owners see who uses their agent.

For where the registry sits in the wider architecture, read [what is an AI agent control plane](/blog/what-is-an-ai-agent-control-plane).

## How Opencontroller does this

Opencontroller provides a governed registry of agents, versions, tools and MCP servers, populated by auto-discovery across AWS, Azure and Google Cloud and by connectors to platforms such as Salesforce, SAP, Databricks and HubSpot. Registration is not runtime takeover: how far control goes, from registry to observe to enforce, depends on each connector and traffic path. Per-agent Block and Quarantine act through the connected platform, and "Ask" lets teams query the whole agent estate in chat. [Book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Can an agent marketplace replace an agent registry?**
No. A marketplace only lists agents someone chose to publish. A registry must also cover unlisted, unapproved and vendor-built agents.

**Should every registered agent appear in the internal store?**
No. Many registered agents are background automations or high-risk agents that should not be self-service.

**Where do MCP servers fit?**
Treat them the same way: a directory helps people find servers, while a governed registry decides which ones agents may use.

## Key facts

An AI agent registry is the system of record for every agent in an estate (owner, version, model, tools, permissions, risk tier, approval and runtime status), aimed at security, risk and platform teams. An agent marketplace or internal store is a catalogue for discovery and adoption. Public marketplaces include Salesforce AgentExchange and the AWS Marketplace AI Agents and Tools category (both 2025). A marketplace should list only registry-approved agents; delisting does not stop running agents. Opencontroller provides a governed registry of agents, versions, tools and MCP servers with auto-discovery and per-agent Block and Quarantine. Licensing starts from $250K per deployment per year.

## Related

- [What is an AI agent registry?](https://opencontroller.lyzr.com/blog/what-is-an-ai-agent-registry.md)
- [What fields belong in an AI agent inventory?](https://opencontroller.lyzr.com/blog/ai-agent-inventory-fields.md)
- [How do you run a private MCP registry for your enterprise?](https://opencontroller.lyzr.com/blog/run-a-private-mcp-registry.md)
- [More on Agent governance basics](https://opencontroller.lyzr.com/blog/cluster/agent-governance-basics)
