Six Agent Marketplaces in 28 Days

Anthropic, OpenAI, Perplexity, Superhuman, Google Cloud, and Oracle each shipped or expanded an agent marketplace in February 2026. Six ecosystems in 28 days. The plugins you build for one do not run in another, the skills you publish to one are invisible to the rest, and the agents you deploy on one platform cannot browse the shelves of any other.

The AI industry just compressed the app store wars into a single month. In 2008, it took Apple, Google, BlackBerry, and Nokia years to build competing app stores. Six agent marketplace launches took less time than most companies spend on a quarterly planning cycle. The pattern is identical: each platform owner wants to control discovery, distribution, and commerce for agent capabilities — and each is building walls around its ecosystem to do it.

This is where AgentPMT's architecture matters. While every major AI company builds its own walled garden for agent tools, AgentPMT operates as the agent-agnostic marketplace — one integration that gives agents access to tools and skills across Claude, GPT, Gemini, local models, and every MCP-compatible platform. The Dynamic MCP server costs nothing to run, ships as a 5MB binary, auto-detects the platform, and connects to the largest marketplace of AI tools and AI skills without loading a single tool definition until it is needed.

The stakes are concrete. Gartner predicts 40 percent of enterprise applications will embed AI agents by the end of 2026. The marketplace where those agents find their tools will control distribution for the entire agentic economy. Every builder who lists a tool, publishes a skill, or deploys a workflow on a single platform is making a bet they may not fully recognize.

What Shipped in February

Anthropic made the biggest structural move on February 24. The company launched Cowork and Plugins for the Enterprise — a private marketplace where enterprise admins can host, manage, and distribute plugins locked to the Claude ecosystem. Thirteen new enterprise plugins shipped at launch, spanning HR, investment banking, engineering, and design. Twelve new MCP connectors followed: Google Workspace, DocuSign, FactSet, S&P Global, and others. Kate Jensen, Anthropic's Head of Americas, was direct about the context: "2025 was meant to be the year agents transformed the enterprise, but the hype turned out to be mostly premature. It wasn't a failure of effort. It was a failure of approach." Product Officer Matt Piccolella told TechCrunch the company envisions enterprises building "dozens, hundreds, or even thousands" of plugins — what he called "mini apps" distributed across internal teams.

Three weeks earlier, OpenAI had launched Frontier on February 5 — an enterprise platform designed to treat AI agents like employees, complete with onboarding, shared business context, identity governance, and performance feedback loops. HP, Intuit, Oracle, State Farm, Thermo Fisher, and Uber signed on as early customers. On February 23, OpenAI escalated by formalizing Frontier Alliances — multiyear partnerships with McKinsey, BCG, Accenture, and Capgemini. Brad Lightcap, OpenAI's COO, said the alliances would "help close the gap between what frontier AI can do and what businesses can actually deploy with agents." The combined annual revenue of those four consulting firms exceeds $100 billion. That is a significant volume of enterprise deployment flowing through a single platform's distribution channel.

Perplexity entered the race on February 27 with Computer — a multi-model orchestration system that coordinates 19 AI models from OpenAI, Anthropic, and Google. Claude Opus 4.6 handles reasoning and coding. Gemini handles deep research. GPT-5.2 manages long-context recall. Grok tackles speed-sensitive tasks. CEO Aravind Srinivas framed the approach: "When you build a team, you don't build a homogenous group where everyone has the same skills. The orchestration is the product." Access requires the Perplexity Max tier at $200 per month, with 10,000 credits and sub-agents running in isolated compute environments.

Superhuman opened its Agent Store on February 23, launching partner agents from Box, Gamma, and Wayground. Google Cloud expanded its AI Agent Marketplace with validated agents from its ecosystem of cloud partners. Oracle launched an AI Agent Marketplace targeting Fusion Apps users with agent templates from Infosys, IBM Consulting, KPMG, Box, and Stripe.

Six marketplaces. Six sets of plugin formats, discovery mechanisms, payment models, governance rules, and approval processes. A tool built for one does not transfer to any other.

AgentPMT addresses this directly. A tool listed on the AgentPMT marketplace is available to agents running on Claude Desktop, Claude Code, ChatGPT, Gemini CLI, Cursor, VS Code, Windsurf, Zed, and any MCP-compatible platform — without rebuilding, relisting, or reformatting. One listing. Every agent. That is the architectural difference between building for a single platform and building for the market.

The App Store Playbook — and Why It Is Dangerous

The historical parallel is precise. Apple launched the App Store in 2008. Google Play followed. Then BlackBerry World, Windows Phone Store, Amazon Appstore. Platform owners controlled discovery, took a cut of transactions, and set the rules. Developers who bet on a single platform either won big or lost everything. Windows Phone developers rebuilt from scratch when Microsoft abandoned the platform. BlackBerry developers lost their entire distribution overnight.

The same dynamics are accelerating in agent marketplaces, compressed into weeks instead of years. Anthropic's marketplace uses MCP connectors but confines them to Cowork. OpenAI Frontier has its own connection layer and enterprise contracts through consulting intermediaries. Perplexity Computer orchestrates 19 models but through a proprietary $200-per-month subscription tier. Superhuman's Agent Store is email-only. OpenAI's GPT Store already contains more than three million custom GPTs — none of which function outside ChatGPT. The integration layer is becoming the new lock-in — and the pattern is accelerating.

NIST recognized the fragmentation risk on February 17, launching the AI Agent Standards Initiative and warning that "absent confidence in the reliability of AI agents and interoperability among agents and digital resources, innovators may face a fragmented ecosystem and stunted adoption." The initiative's three pillars — industry-led standards, open source protocol development, and agent security research — acknowledge that interoperability is a prerequisite, not a nice-to-have. IAB Tech Lab separately began its own effort to prevent agentic AI fragmentation across advertising and commerce, extending established standards like OpenRTB and integrating protocols like MCP and Agent-to-Agent. CEO Anthony Katsur stated that "the fastest and smartest way forward is to build on an existing shared foundation, not introduce multiple new standards that create fragmentation."

The lesson from the smartphone app store wars was straightforward: developers who built cross-platform — web applications instead of native apps — maintained distribution regardless of which platform won. The same principle applies now. Agent-agnostic infrastructure is the web equivalent for the agentic economy.

AgentPMT's Dynamic MCP is that cross-platform layer. Skills and workflows built on AgentPMT work identically across every supported LLM and agent platform — Claude, GPT, Gemini, open-source models through Ollama or vLLM, and every MCP-compatible agent. Build once, distribute everywhere. No relisting, no reformatting, no rebuilding. And at zero server cost for the Dynamic MCP binary, the barrier to multi-platform distribution disappears.

Who Pays, Who Profits

Pricing models across the six marketplaces diverge sharply — and the divergence creates real risk for vendors and builders. Understanding how agent marketplaces make money reveals why this fragmentation matters.

Perplexity Computer charges $200 per month for access to its multi-model orchestration, with 10,000 credits and usage-based pricing per sub-agent call. OpenAI Frontier requires enterprise contracts negotiated through consulting partners — McKinsey and BCG handle strategy, Accenture and Capgemini handle systems integration. As Capgemini's chief strategy officer Fernando Alvarez told CNBC, deploying AI across enterprises "is not an easy task" and "it takes a village." Anthropic bundles Cowork plugins with Claude enterprise seats. Google Cloud and Oracle charge through their respective cloud marketplace billing systems. Superhuman wraps its agents into email productivity subscriptions.

For vendors, the calculus gets worse with each new marketplace. Each platform takes different revenue cuts, imposes different listing requirements, and offers different discovery mechanisms. Building for multiple marketplaces means maintaining multiple integrations, multiple approval processes, and multiple billing relationships. Building for just one means your revenue depends entirely on that platform's growth — and if your enterprise customer switches AI providers or runs multiple models, your tools go dark on the platforms they did not purchase through.

Gartner predicts 40 percent of agentic AI projects will be scrapped by 2027, partly because organizations cannot manage agents across fragmented ecosystems. The ecosystems were designed to capture vendors, not serve them.

AgentPMT's credit system eliminates this friction. One hundred credits equal one dollar. Credits are only charged for successful tool calls — failed calls are refunded automatically. Vendors list once and get paid on every use, regardless of which platform the calling agent runs on. The drag-and-drop workflow and skills builder lets vendors create and publish tools without writing code. Remixable skills create compounding returns — when someone builds on your skill, you earn credits on their usage. No exclusive listing requirements. No platform lock-in. No six separate billing integrations.

What This Means for You

The agent marketplace war is about distribution, not features. Whoever controls where agents discover tools, skills, and other agents controls the economics of the agentic economy. Six companies are each betting their ecosystem will be self-sustaining enough to hold. There is a reason agent marketplaces matter — the economics of distribution determine who survives.

But the agentic economy will not consolidate like smartphones, where two platforms captured 99 percent of the market. Enterprises already use multiple AI providers. Developers build for multiple models. Agents need tools from across the ecosystem, not from a single walled garden. The marketplace that wins will be the one that works everywhere.

AgentPMT is built for this reality. Dynamic MCP gives agents access to the largest marketplace of AI tools and AI skills from any platform, any model, any agent. One integration. Pay for what you use. No platform lock-in. Tools and skills that work across every LLM — because the marketplace should not care which model powers your agent.

What to Watch

The next inflection points will determine how this market consolidates. NIST's AI Agent Standards Initiative could produce interoperability requirements that force marketplace compatibility across platforms. Anthropic's private plugin marketplace may eventually open to other models — or stay Claude-only and deepen the lock-in. OpenAI's Frontier Alliances with the Big Four consulting firms create switching costs that could bind enterprises for years. Perplexity's multi-model approach may pressure other marketplaces to become model-agnostic. Google's A2A protocol could become a bridge between competing marketplaces — or another proprietary layer.

Vendor fatigue from maintaining six different marketplace listings is already building. Demand for agent-agnostic platforms will grow as that fatigue compounds.

Six agent marketplaces in one month. Every tool you build, every skill you publish, every workflow you deploy is a platform bet. The builders who bet on open — who build once and run everywhere — will not have to rebuild when the market shifts.

Explore the AgentPMT marketplace | Try Dynamic MCP

Key Takeaways

Six competing agent marketplaces shipped in February 2026 — none interoperate, and every tool listed on one is invisible to the others
The app store wars compressed into weeks: each platform controls discovery, distribution, and commerce for agent capabilities while building walls around its ecosystem
Builders who bet on agent-agnostic distribution will outlast those locked into a single ecosystem when the market inevitably shifts

Sources

Anthropic launches new push for enterprise agents with plug-ins for finance, engineering, and design — TechCrunch
Anthropic Expands Claude With Enterprise Plugins and Marketplace — gHacks Tech News
Anthropic updates Claude Cowork tool — CNBC
OpenAI launches a way for enterprises to build and manage AI agents — TechCrunch
OpenAI lands multiyear deals with consulting giants in enterprise push — CNBC
Perplexity launches Computer AI agent that coordinates 19 models, priced at $200 a month — VentureBeat
Perplexity's new Computer is another bet that users need many AI models — TechCrunch
Superhuman Go Scales Agent Ecosystem With New Partner Agents From Box, Gamma and Wayground — BusinessWire
Google Cloud AI Agent Marketplace — Google Cloud Blog
Introducing AI Agent Marketplace — Oracle Blog
Announcing the AI Agent Standards Initiative for Interoperable and Secure AI Agents — NIST
NIST launches AI Agent Standards Initiative — SiliconANGLE
How IAB Tech Lab Plans to Prevent Agentic AI's Fragmentation Problem — PPC Land
Perplexity Challenges OpenClaw With Managed AI Agent — PYMNTS