AI Agent Marketplaces: The Next App Store?

An AI agent marketplace is a catalogue where businesses and consumers find, buy and deploy AI agents and the tools they use. In 2026 the real buying happens in cloud storefronts from AWS, Google, Microsoft and Salesforce, while consumer stores inside ChatGPT keep changing. None yet works like Apple's App Store.

This guide maps where agents are actually bought and sold in 2026, what marketplaces charge, how agents are priced, why "listed" doesn't mean "safe", which standards make agents portable, and what it all means for businesses in the UAE — including property companies deciding whether to build, buy or list.

Key takeaways

• Enterprise storefronts are where agents are bought. AWS, Google, Microsoft, Salesforce and Anthropic all sell agents through marketplaces tied to cloud contracts, with seller fees around 3%.
• Consumer stores keep resetting. OpenAI is retiring custom GPTs, has folded apps into a Plugin directory and allows commerce for physical goods only; builders get no revenue share.
• Pricing is shifting to usage and outcomes. Fin charges $0.99 per outcome; Salesforce sells credits and per-conversation pricing alongside seats.
• Listing isn't vetting. The official MCP Registry keeps servers with known vulnerabilities, and a 2026 study found a quarter of agent skills had at least one security flaw.
• The plumbing is standardising. A2A, MCP and AGENTS.md now sit together under the Linux Foundation's Agentic AI Foundation, which makes agents more portable between stores.

What is an AI agent marketplace?

An AI agent marketplace is a catalogue — usually run by a cloud or software platform — where buyers discover, purchase, deploy and govern AI agents, or the tools and skills agents use, with billing and permissions handled by the platform.

Definition

AI agent marketplace — a platform catalogue for finding, buying, installing and managing AI agents and agent tools. No standards body defines the term; in practice it covers enterprise storefronts, consumer directories inside AI assistants, and open registries of agent tools.

Three layers are emerging, and they work very differently:

Layer	Examples	Who buys	How money flows
Enterprise storefronts	AWS Marketplace AI agents and tools; Google's Agent Marketplace in Gemini Enterprise; Microsoft Agent Store and Marketplace; Salesforce AgentExchange; Claude Marketplace	IT and procurement teams	Contracts billed through the cloud provider, often against committed spend
Consumer directories	ChatGPT's Plugin directory; Claude's connectors and skills; Gemini connected apps	Individuals and teams	Mostly free to list; businesses monetise on their own sites; platforms sell ads
Open registries	The MCP Registry; A2A agent cards; skill registries	Developers and other agents	None — discovery only

Is an AI agent marketplace the next App Store?

Not yet. The App Store works because apps are self-contained, sandboxed and paid for through one checkout; agents need access to your data and systems, are chosen by IT rather than tapped on a home screen, and are increasingly called by other agents.

For scale: Apple says its App Store ecosystem facilitated $1.4 trillion in developer billings and sales in 2025. No agent marketplace publishes anything comparable, and none shares the App Store's core ingredients:

	App Store	AI agent marketplaces, 2026
Buyer	Consumer, one tap	IT or procurement, often with security review
What you install	A sandboxed app	Software that acts with your permissions and data
Payment	One checkout, platform commission	Cloud contracts and commitments; ads in consumer AI
Review	Platform reviews every app	Varies from light-touch to none for open registries
Discovery	Rankings and search	Metadata that an assistant or agent reads, plus paid placement
Builder income	Revenue from sales	Little from consumer directories; enterprise sales through clouds

The closest thing to App Store dynamics is inside the assistants: OpenAI now makes money from the audience — ChatGPT's ads reached a $1 billion annualised run rate in under 200 days, and in September 2026 it began testing "Sponsored Agents", letting users chat with a business's agent after clicking an ad.

Where can businesses buy AI agents in 2026?

Businesses buy agents mainly through the big cloud and software platforms, which bundle procurement, billing and governance with the catalogue.

Platform	Marketplace	What's notable	Seller terms
AWS	AI agents and tools category in AWS Marketplace (July 2025)	Conversational "agent mode" for discovery; a private AWS Agent Registry, generally available since August 2026, for governing agents in-house	3% on SaaS listings; 20% on public server-based listings
Google Cloud	Agent Marketplace and Agent Gallery in Gemini Enterprise (April 2026)	Partners sell agents directly to enterprise users; a $750 million partner fund	Vendors keep 97% on standard offers; commit drawdown capped at 25% of the commitment
Microsoft	Agent Store in Microsoft 365 Copilot; Microsoft Marketplace (September 2025)	More than 3,000 AI apps and agents at launch; Agent 365 for governing agents, $15 per user per month since May 2026	Microsoft keeps 3% of the licence cost on transactable offers
Salesforce	AgentExchange (March 2025)	Launched with more than 200 partners; agents run inside Agentforce	Salesforce's own pricing mixes credits, conversations and seats
Anthropic	Claude Marketplace (relaunched September 2026)	More than 2,000 connectors and plugins; customers can use part of their committed Anthropic spend on partner products	Partner products billed through Anthropic

The commercial draw isn't discovery — it's procurement. A buyer who has already committed spend with a cloud provider can often buy a partner's agent against that commitment, with one invoice and existing security terms. Google caps that drawdown at 25% of a commitment; Anthropic lets customers apply "a portion" of committed spend. For a vendor, being in the marketplace can shorten a sales cycle by months.

What happened to consumer agent stores like the GPT Store?

They've been reset repeatedly. OpenAI launched the GPT Store in January 2024 with more than 3 million custom GPTs and promised a builder revenue programme; that programme stayed limited to a select group of US builders, and custom GPTs are now being retired.

• GPTs to plugins. OpenAI launched apps in ChatGPT in October 2025, then in July 2026 turned its app directory into a Plugin directory shared by ChatGPT and Codex. For Enterprise workspaces, custom GPTs are scheduled to stop running on 11 December 2026, and custom actions don't carry over automatically.
• Commerce limits. Plugins may conduct commerce only for physical goods — no digital products, subscriptions, tokens or credits.
• Monetisation moved to ads. The builder gets distribution; OpenAI earns from advertising and, soon, sponsored agents.
• Anthropic and Google. Claude has a connectors directory and an open Agent Skills format published in December 2025. Google shares custom Gems by link and adds partner apps to Gemini in curated waves.

Common misconception

"You can build a business selling GPTs." Payouts never went beyond a closed pilot, the GPT format itself is being retired, and plugins can't sell digital goods or subscriptions. Treat consumer directories as distribution channels that send people to your own product — the way DEWA and Bayut use ChatGPT — not as app-store revenue.

How are AI agents priced?

Agents are priced per user, per use or per outcome — and many vendors now offer all three at once.

Model	Example	Good for	Watch out for
Per user	Microsoft Agent 365, $15 per user per month	Predictable budgets; broad internal use	Paying for seats that don't use the agent
Per use	Salesforce Flex Credits, $500 per 100,000 credits; $2 per conversation	Variable workloads	Bills that scale faster than value; opaque credit maths
Per outcome	Fin AI Agent, $0.99 per outcome	Customer service with clear resolution metrics	Disputes over what counts as an outcome
Marketplace fee (seller side)	AWS 3% on SaaS; Microsoft 3%; Google vendors keep 97%	Reaching enterprise buyers	Less relevant for small or consumer products

Outcome pricing is the trend to watch: it aligns cost with value, which is exactly what buyers struggling with enterprise AI ROI want. Salesforce agreed in June 2026 to acquire Fin — the AI agent business formerly known as Intercom — for about $3.6 billion.

Are AI agents from marketplaces safe?

Not automatically. Most catalogues check who published a listing, not what the code does, and agent tools run with the permissions you give them.

• Registries are permissive by design. The official MCP Registry's moderation policy says it removes only illegal content, malware, spam and completely broken servers — servers with known vulnerabilities can stay listed.
• Supply-chain attacks are real. In September 2025 Postmark warned that a fake "postmark-mcp" npm package, after 15 clean versions, added a backdoor in version 1.0.16 that secretly copied every email it sent to an outside server.
• Skill stores have been seeded with malware. Security firm Koi reported 341 malicious skills among 2,857 on ClawHub, the skill registry of the OpenClaw personal agent, in February 2026; OpenClaw then began scanning all skills with VirusTotal.
• Flaws are common. An academic study of 31,132 agent skills from two marketplaces found 26.1% had at least one vulnerability and 5.2% showed high-severity patterns suggesting malicious intent.
• The risk list is now standard. OWASP's Top 10 for Agentic Applications (December 2025) names agentic supply-chain vulnerabilities, tool misuse and identity and privilege abuse among the main threats.

Before installing any agent or tool, ask:

1. Who published it, and is the publisher verified?
2. What permissions does it need — read, write, send, pay — and can you narrow them?
3. Where does it send data, and is that covered by your data-protection obligations?
4. Can you test it in a sandbox with dummy data first?
5. Is it pinned to a version, so it can't silently change what it does after approval?
6. Does it log actions you can audit — the controls in the AI governance guide?

Which standards make AI agents portable between marketplaces?

A small set of open standards — MCP for tools, A2A for agent-to-agent communication and shared formats for skills and instructions — now sit under one neutral home, which lets an agent built for one store work in another.

• MCP (Model Context Protocol) connects agents to tools and data; its official registry, still in preview, lists public MCP servers.
• A2A (Agent2Agent) lets agents discover and call each other; since v1.0 an agent publishes a signed "agent card" describing its skills and authentication. A2A joined the Linux Foundation's Agentic AI Foundation in August 2026.
• Agent Skills packages instructions and scripts an agent loads when needed; Anthropic published it as an open standard in December 2025.
• AGENTS.md gives coding agents a standard brief for working in a codebase — this site has one.

Portability matters for buyers: an agent built on open standards can be moved when a marketplace changes its rules, as OpenAI's did twice in two years. The agentic AI guide explains how MCP and A2A moved into neutral governance, and the agent-to-agent transactions guide covers the payment and identity layers on top.

What does the latest data show?

Forecasts point to fast growth in agent software, while the evidence on marketplaces themselves is still thin and mostly vendor-reported.

What the data shows

• Spending (forecast, September 2026): worldwide spending on AI agents and assistants of $16.5 billion in 2025, $29.2 billion in 2026 and $65.5 billion in 2027. — Gartner
• Adoption (forecast, August 2025): 40% of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5% in 2025. — Gartner
• Exchanges (forecast, October 2025): by 2028, more than $15 trillion of B2B spend will flow through "AI agent exchanges". — Gartner
• Catalogue size (vendor figures): more than 3,000 AI apps and agents at Microsoft Marketplace's launch; more than 2,000 connectors and plugins in the Claude Marketplace. — Microsoft, Anthropic
• Consumer AI advertising (vendor figure, August 2026): ChatGPT Ads reached a $1 billion annualised revenue run rate in under 200 days. — OpenAI

One experiment shows why agent marketplaces need care. In Anthropic's Project Deal, Claude agents negotiated 186 deals worth just over $4,000 for employees in a classified-ads market. People represented by stronger models got objectively better deals — and those with weaker models didn't notice their disadvantage. As agents start buying from agents, the quality of your agent becomes a commercial variable.

What does this mean for the UAE?

There's no UAE government agent marketplace; instead, Dubai certifies trusted AI suppliers, and UAE organisations reach users through global agent directories.

• Certification over catalogues. The Dubai Centre for Artificial Intelligence runs the Dubai AI Seal; 325 companies had applied by May 2025. In October 2025, Dubai Finance urged government entities to restrict AI service and consultancy contracts to Seal-certified companies — effectively an approved-supplier list for public-sector AI.
• Policy push. The UAE Cabinet's April 2026 framework targets agentic AI across half of government sectors within two years, which will pull agent vendors toward the Seal and government procurement.
• Global directories as distribution. DEWA launched its services in ChatGPT's app directory in February 2026, and Bayut launched a property-search app on ChatGPT in June 2026.
• Buying through the clouds. UAE enterprises buy agents the same way as elsewhere — through AWS, Google, Microsoft and Salesforce marketplaces — with data-residency and procurement rules applied at the contract level.

What should property businesses do about agent marketplaces?

Real estate is already listed in the consumer directories, so the decision for developers and brokerages is whether to list, buy or build.

Zillow launched an app in ChatGPT in October 2025, Redfin followed in February 2026 and Bayut in June 2026; apartments.com joined Gemini's connected apps in September 2026; property-management platforms such as AppFolio and Yardi offer connectors in Claude's directory. That means a buyer can search listings inside an assistant before ever visiting a portal or a developer's website.

Business	List	Buy	Build
Portal or large developer	An app in ChatGPT or a Gemini connected app, fed by clean listing data	Agent tooling through your cloud marketplace	Your own agents for search, qualification and bookings
Brokerage	Make sure your listings are complete on the portals that are listed	CRM, WhatsApp and voice agents from established vendors	Custom workflows on top — see the agentic real estate CRM guide
Property manager	Connectors that let assistants read your systems safely	Tenant-service and maintenance agents	Rarely worth it before the basics work

For most brokerages, the right move in 2026 is to be well represented in the directories through the portals, buy proven agents for lead handling, and build only where the workflow is genuinely unique — the pattern described in AI in real estate.

What is likely to happen next?

Expect consolidation of consumer directories, paid placement inside assistants, stronger vetting, and marketplaces where agents buy tools and services from other agents.

• Paid distribution grows. Sponsored agents and ads make assistant visibility something businesses can buy, not only earn — AI search and GEO become part of marketing budgets.
• Vetting tightens. After the malware incidents, expect scanning, verified publishers and version pinning to become standard across registries.
• Agents become the buyers. AWS already lets agents pay for APIs, and Gartner expects trillions in B2B spend through agent exchanges by 2028 — the direction covered in machine customers.
• Outcome pricing spreads. Vendors that can measure outcomes will price on them; buyers will demand it.

Final takeaway

AI agent marketplaces are real, but they're not the App Store — yet. Businesses buy agents through cloud storefronts that win on procurement, not discovery; consumer directories inside ChatGPT, Claude and Gemini are distribution channels whose rules keep changing; and open standards underneath are what keep agents portable. Buy through marketplaces for speed and governance, list in assistant directories for reach, vet every agent as you would a new supplier, and don't build a business model on any single store's current rules.