Anthropic turns Claude into a marketplace with more than 2,000 plugins and connectors
Anthropic opened a Claude Marketplace on 27 September 2026, gathering plugins, connectors, agents, and consulting partners, with more than 2,000 integrations live at launch. Vendors see a distribution channel, developers a reason to learn MCP and Agent Skills.
27 September 2026. Anthropic opens the Claude Marketplace to the public. More than 2,000. That is the number of plugins and connectors already live at launch, from Atlassian, Google, Microsoft, Notion, and Salesforce. MCP. The Model Context Protocol becomes the technical gateway for anyone who wants to publish on the platform. Why it matters: Claude stops being a mere model and becomes a marketplace, and that is a change in kind, not degree.
From a model to a distribution platform
Until now, Claude was sold as an assistant, a subscription, an API endpoint. The announcement on 27 September 2026 changes the category. Anthropic is no longer offering just a language model but a commercial ecosystem: a single place to discover plugins, connectors, ready-made agents, and even consulting partners for deploying the technology across an organization.
The launch catalog signals the scale of the ambition. On the integration side, there are connectors supplied by Atlassian, Google, Microsoft, Notion, and Salesforce. On the product side, agents and software powered by Claude are sold by CrowdStrike, Cursor, Harvey, Legora, Lovable, and Snowflake. On the services side, firms such as Accenture, Boston Consulting Group, and Deloitte handle large-scale rollout.
The comparison Anthropic itself reaches for is an app store, a Google Play Store applied to AI. The stated goal is to help teams discover products that already work with Claude, while giving developers and partners a direct route to the platform’s existing customers.
The bet on avoiding OpenAI’s app-store failure
The initiative has a precedent, and that is precisely what makes it interesting. OpenAI tried an app marketplace that never really took off, short on depth and openness. Anthropic claims to have learned from that failure by opening the door to any developer, rather than to a handful of curated partners.
The technical machinery rests on two building blocks. The Model Context Protocol, or MCP, standardizes how a model reaches external tools and data. Agent Skills provide a format for packaging reusable capabilities. A developer who builds a connector on MCP or an agent skill can list the work on the marketplace, where Claude’s enterprise customers can discover and buy it.
That openness has a direct strategic consequence. By becoming a distribution channel, Anthropic no longer sells only its own capabilities but captures value created by an entire ecosystem of third parties. It is the shift from model vendor to market infrastructure, the same move that made classic software marketplaces powerful.
What it changes for teams
For an enterprise, the marketplace changes the adoption calculus. Instead of evaluating a bare model, the buyer evaluates an ecosystem: who has already connected, which connectors cover its internal tools, which partners know how to deploy. The presence of Salesforce, Notion, and Microsoft at launch lowers the perceived integration risk, and the presence of firms like Accenture lowers the deployment risk.
The downside is lock-in. The more connectors and agents an organization wires onto Claude, the costlier a switch to another model becomes, even if the MCP layer is theoretically portable. The marketplace is also a retention mechanism: value migrates from the model to the integrations that surround it.
For security teams, the announcement adds surface area to watch. Every connector is a potential path into company data, and every third-party agent is code running with real permissions. The presence of CrowdStrike or Harvey in the catalog does not settle the underlying question: who audits what enters the company’s Claude environment.
For procurement and IT leadership, the marketplace turns a technical choice into a sourcing decision. Buyers who once compared models on benchmark scores now compare catalogs on connector coverage, partner maturity, and auditability. That shift favors the vendor with the deepest enterprise relationships, which is exactly what the presence of Accenture, BCG, and Deloitte signals. The model becomes a commodity; the network around it becomes the moat.
The MCP standard becomes a power struggle
The most consequential element is not the catalog but the standard underneath it. The Model Context Protocol is becoming the common language for connecting models to tools, adopted well beyond Anthropic’s own ecosystem. By making MCP the currency of its marketplace, Anthropic promotes a standard that benefits everyone but from which it takes the first dividend.
A developer who learns MCP and Agent Skills is not investing only in Claude: they acquire a portable skill usable on other platforms that adopt the same protocol. That is the argument that makes the openness credible and sets this announcement apart from a mere storefront. The marketplace sells products, but first it promotes a standard.
The battle now opening is therefore twofold: a commercial battle over the distribution of agents, and a technical battle over the protocol that connects them. Both play out on the same ground, and Anthropic has just placed a major piece there.
The competitive response and the risk of fragmentation
Anthropic’s announcement lands at a moment when competition is decided less by model size than by the ability to connect models to things. OpenAI has already failed once to sustain an app marketplace, but has not dropped the idea: its agents and the GPT Store have mostly served as a cautionary tale. Google, with Gemini, and Microsoft, with Copilot, are meanwhile pushing native integrations deep inside their office suites, where the model disappears into tools that are already deployed.
The risk for the user is fragmentation. If every vendor imposes its own protocol and its own catalog, enterprises will have to choose an ecosystem rather than a model, and connectors written for one will not carry over to another. That is exactly what the Model Context Protocol tries to prevent: a single standard for connecting models to tools, which would make integrations portable from one platform to the next.
Anthropic is therefore playing a double game. On one side, it builds a catalog that locks in customers and captures third-party value. On the other, it promotes an open standard that makes that catalog less indispensable in the long run. It is a delicate but rational position: if MCP wins, Anthropic remains credible as the lab that launched it, even if other platforms benefit. Openness here is not generosity: it is insurance against the failure of the closed platform.
Verdict
If you run a team already using Claude, treat the marketplace as an adoption lever and a risk at once: adopt the connectors that cover your tools, but demand an inventory and a permission audit before letting third-party agents reach your data. If you are a developer, the priority is to learn MCP and Agent Skills, because that is where the portable value of the announcement lies, regardless of which platform wins. If you watch the market, note the trajectory: Claude is becoming a distribution channel, and the next logical step is competition on the depth of catalogs rather than the size of models. The race is no longer only about who speaks best, but about who connects the most.