An AI agent marketplace is a storefront where you browse, install, and pay for AI agents someone else already built, the same way an app store works for mobile apps. In 2026 there isn’t one dominant marketplace. There are three different kinds, built by three different groups of companies, and they solve different problems. This guide covers what’s actually live, what each kind is good for, and when installing someone else’s agent is the wrong move entirely. We build Shogo, an open-source AI agent you describe agents to rather than shop for, so we’ve also been specific below about where Shogo is a better route than a marketplace and where it isn’t.
The three kinds of AI agent marketplace that actually exist
1. CRM-native marketplaces
These live inside a platform you probably already pay for, and the agents only work inside that platform.
Salesforce AgentExchange, announced in March 2025, is the clearest example. It’s a marketplace for Agentforce, Salesforce’s agent platform, where partners list pre-built agents, apps, and Slack integrations that install directly into your Salesforce org. The pitch is straightforward: discover, try, and buy an agent built by a partner who already knows your CRM’s data model, instead of building one from scratch.
The tradeoff is just as straightforward. An AgentExchange agent is built for Salesforce, priced for Salesforce customers, and won’t help you if your workflow lives outside Salesforce.
2. Cloud infrastructure marketplaces
AWS Marketplace and Google Cloud Marketplace both list AI agent products the way they’d list any other software, including enterprise agent platforms like Agentforce itself. You find them through the same procurement flow you’d use for a database or a monitoring tool, which matters if your company already buys software exclusively through AWS or GCP’s marketplace for billing and approval reasons.
Google Agentspace is a related but different product. It’s Google Cloud’s own agent-driven enterprise platform, with a no-code Agent Assembler for building agents against your company’s data, rather than a storefront for other vendors’ agents.
Microsoft Copilot Studio sits in the same space on the Microsoft side: a low-code platform for building and publishing custom agents into Teams, SharePoint, and Microsoft 365. It meters usage in Copilot Credits rather than per-seat pricing, a prepaid pack runs $200 per month for 25,000 credits, or $0.01 per credit pay-as-you-go. Microsoft also runs Agent 365, a separate governance and identity layer that tracks agents built anywhere in the Copilot Studio ecosystem, which reached general availability on May 1, 2026. That split matters if you’re evaluating Microsoft’s agent stack: governance and usage are billed on different models, not bundled into one number.
3. Protocol and framework registries
This is the newest and most technical category. Model Context Protocol (MCP), the open standard for connecting AI models to external tools and data, has spawned its own registries. Google’s Agent Registry is a centralized catalog for storing, discovering, and governing MCP servers and tools. These aren’t consumer-facing marketplaces with pricing pages. They’re closer to a package registry for developers building agents on top of a framework, similar in spirit to how npm or PyPI work for code libraries.
Hugging Face also hosts agent-related projects through Spaces, its platform for sharing ML demos and apps, though it functions more as an open community directory than a curated commercial marketplace.
What to check before you install someone else’s agent
An agent you didn’t build makes decisions inside your systems using assumptions you didn’t set. Before installing anything from any of the three categories above, worth checking:
-
What data it touches. A CRM-native agent typically only reads and writes within that platform’s data model. A cloud-marketplace agent might request broader permissions, check exactly what it’s asking for.
-
Who maintains it. A marketplace listing doesn’t guarantee ongoing support. Check when it was last updated and whether the publisher is the platform vendor itself or a third party.
-
What happens when it’s wrong. Pre-built agents are tuned for a general case, not your specific workflow. Know what review step exists before an agent-initiated action (a refund, a record update, an email) actually goes out.
-
Whether it’s portable. A CRM-native agent is locked to that CRM. If you switch platforms, the agent doesn’t come with you.
When a marketplace agent is the wrong fit
Marketplaces solve a real problem: you need an agent today and don’t want to build one. But every agent listed in a marketplace was built for the average customer of that platform, not for your specific process. If your workflow is simple and matches what the platform already assumes, a marketplace agent can save real time. If your process has its own rules, exceptions, or data sources that don’t fit neatly into Salesforce, AWS, or Microsoft’s assumptions, you’ll spend more time working around a pre-built agent’s limits than you would have spent describing your actual process to a platform that lets you build the agent yourself.
That’s the gap a builder fills instead of a marketplace: you describe the agent you need in plain English and it’s built around your actual workflow and your actual data, not a generic template.
Where Shogo fits, and where it doesn’t
Shogo is not a marketplace. There’s no catalog of third-party agents to install in one click. It’s an open-source AI agent that builds the agent, tool, or internal app you describe, runs it, and puts it online, inside the same workspace where your team already chats with it.
Where Shogo is the better route than a marketplace:
-
Portability. The marketplace problem above, an agent locked to the platform you bought it from, is the one Shogo is designed around. Shogo is open source and self-hostable, and the pricing page is explicit that the software your agents build stays yours if you cancel.
-
No single-platform assumption. A CRM-native agent only sees that CRM’s data. A Shogo agent works across the tools you connect, and you can switch between OpenAI, Anthropic, and Google models without switching platforms.
-
Predictable cost. Paid plans are unlimited usage within fair-use 5-hour and weekly windows rather than a credit pool: Free at $0, Basic at $8 per month, Pro at $20 per seat per month, and Business at $40 per seat per month with SSO and audit logs.
Where a marketplace is the better route:
-
Deep, single-platform workflows. If your process lives entirely inside Salesforce and you want a partner-built agent that already understands your Salesforce data model, AgentExchange gets you there faster than building one.
-
Procurement rules. If your company can only buy software through AWS or Google Cloud Marketplace, that constraint decides the route before any feature comparison does.
-
Zero setup time. Shogo’s free plan includes starter templates, but describing your own workflow still takes more time than installing a finished listing.
A Shogo agent can also be set up to own a whole role rather than a single task, closer to an AI employee than to an app-store download.
FAQ
Is there an official AI agent marketplace?
No single official one. The closest equivalents are platform-specific marketplaces like Salesforce AgentExchange, cloud marketplaces from AWS and Google Cloud, and open registries for MCP-based tools. Which one applies depends on what platform your agent needs to run inside.
What’s the difference between an AI agent marketplace and an AI agent builder?
A marketplace sells agents someone else already built, ready to install. A builder, like Shogo, lets you create an agent from scratch around your own workflow and data. Marketplaces trade customization for speed; builders trade a bit more setup time for an agent that actually matches your process.
Is Shogo an AI agent marketplace?
No. Shogo is an open-source AI agent that builds the agents and apps you describe, rather than a catalog of pre-built agents from other vendors. It’s self-hostable, works across every major model provider, and has a free plan, with paid plans from $8 per month.
Is AgentExchange only for Salesforce customers?
Yes. AgentExchange agents are built for Agentforce, Salesforce’s agent platform, and install into a Salesforce org. They don’t run outside that environment.
What is an MCP registry?
A catalog for Model Context Protocol servers and tools, the components developers use to connect AI agents to external data and systems. Google’s Agent Registry is one example. These are developer-facing, not consumer storefronts.
Do I need to pay for every agent in a marketplace?
Pricing varies by listing and by platform. Some marketplace agents are free add-ons, others are metered by usage rather than seats, as with Microsoft Copilot Studio’s credit-based pricing. Always check the specific listing’s pricing page rather than assume a flat per-seat cost.
Can I move an agent from one marketplace to a different platform?
Generally no. Marketplace agents are built against a specific platform’s data model and APIs, so they don’t transfer. This is the main reason some teams choose to build a portable, custom agent instead of relying on a marketplace listing.
