Anthropic opens Claude Marketplace with 2,000 connectors: the game is no longer the model, it's who controls distribution

By Zendoric · September 28, 2026.

Anthropic launched Claude Marketplace on September 23, a single catalog inside its platform for adding connectors and plugins, buying agents and products built on Claude, and finding consultancies to help deploy it, according to the company's own blog as reported by gHacks. It went live with more than 2,000 connectors and plugins available from day one.

The catalog is organized into three sections. The first gathers third-party connectors and plugins — Atlassian, Google, Microsoft, Notion and Salesforce are among the launch partners — which allow Claude to read and act on tools a company already uses. The second offers complete agents and products built by partners such as CrowdStrike, Cursor, Harvey, Hebbia, Legora, Lovable and Snowflake; according to Anthropic, these products can be paid for out of the spend an organization has already committed to the company. The third lists consultancies and integrators — Accenture, Boston Consulting Group and Deloitte, among others — for those who want help deploying Claude across the organization.

There are three ways into the catalog, according to Anthropic's documentation: developers build connectors or plugins with MCP (Model Context Protocol, the open standard that allows an AI model to be connected to external tools and data) and Agent Skills, and submit them through the official documentation; companies selling Claude-based products request a listing via an application process; and consultancies join the Claude Partner Network through a dedicated form. Anthropic has not published how it reviews each submission, how long approval takes, or whether individual Pro and Max subscribers will have the same purchasing options as enterprise accounts.

The launch comes shortly after Anthropic offered up to $250 in free Claude Code cloud credits to its Pro and Max subscribers, as part of a broader expansion of what paying customers receive. Broadly speaking, no major AI lab has yet solved the question of how to monetize beyond the per-token subscription, and building a third-party marketplace — with a revenue split still undisclosed — is an obvious answer, one OpenAI already tried with its GPT store.

Our reading is that this move matters less for what it adds to Claude today than for what it reveals about where competition between labs is heading. When the quality of frontier models converges — Anthropic, OpenAI and open Chinese models compete ever more closely on hard benchmarks — the decisive advantage stops being the model alone and shifts to distribution: who is already plugged into the tool the customer uses, who bills against an already-committed budget, who has Deloitte or Accenture selling the rollout door to door. It is the same underlying dynamic as always: the war is fought in the plumbing — connectors, agent standards, enterprise sales channels — not just on the capability scoreboard.

The "committed spend" payment design deserves a close reading. It is convenient for the customer, who does not open a new billing line for every product, but it is also an elegant way of tying that spend to Anthropic: the more you buy inside the marketplace, the deeper the organization sits within its ecosystem, and the more expensive it becomes to migrate to another provider later. That is not necessarily a bad thing — Apple's App Store and Salesforce's AppExchange work the same way — but it is worth calling it by its name: it is a retention strategy, not just a convenience.

In the short term, the lack of transparency about how connectors are reviewed is the point an IT administrator should watch before any other. Every listed connector gives Claude access to an external service — email, the CRM, the document manager — and Anthropic has not explained what security or privacy controls it requires from outside developers before approving their inclusion. With 2,000 integrations from launch day, the risk surface is growing far faster than the public information about how it is audited.

Over the longer term, if MCP consolidates as the de facto standard for connecting agents to enterprise tools — which is starting to look likely, given that other labs are already adopting it — marketplaces like this genuinely reduce the friction for a midsize company to deploy AI across its processes without an in-house engineering team. That is precisely the kind of boring but necessary infrastructure that, added up across thousands of organizations, ends up moving the aggregate productivity needle more than any new model announcement. The risk is that this infrastructure ends up concentrated in two or three closed platforms; the opportunity is that, properly governed, it becomes the channel through which the abundance AI promises reaches the ordinary company faster, not only those that already had their own engineering team to integrate it by hand.