Anthropic opens Claude plugin publishing to developers
Anthropic opened a portal on 25 September for developers on paid Claude plans to submit plugins to its directory, follow the review and see usage analytics after publication. A plugin can consist of a remote Model Context Protocol connector or a bundle combining MCP servers and Agent Skills; Claude Code bundles can also include language-server features, commands, hooks and agents. The portal automatically validates and safety-scans submissions, gives feedback, and lets approved developers choose when to publish. The move makes it easier for outside software makers to distribute capabilities inside Claude rather than ask each customer to configure a connector manually. Anthropic says millions of people already connect Claude to apps, work tools and data.
The developer experience includes data that can shape a product decision. Once a plugin is live, its maker can see installs by Claude product surface and version, directory views and the searches that lead users to the listing. Those measures help identify whether a listing is discoverable, whether a change produces adoption and where to invest support. Anthropic says developers can ‘review status and feedback’ from its safety scan before publication. A developer can submit a single remote MCP connector without creating a larger bundle, or put skills and servers in a GitHub repository. That lower-friction entry point may increase the number of specialized integrations available to customers.
Anthropic said Claude supports a newer MCP specification with a stateless core and extensions for interactive interfaces within chat and enterprise-managed authentication. The latter matters when an organization wants employees to use a vetted tool without each person completing a separate OAuth configuration. The company described plugins as its main route for third-party extensions, while retaining skills and connectors as building blocks. It plans a more unified discovery experience across Claude and Claude Code over the following weeks. Developers with existing directory listings need not change them immediately. That staged migration acknowledges a practical platform constraint: a better submission path only succeeds if existing integrations remain usable while the catalog grows.
The announcement follows the 23 September Claude Marketplace, which gathers connectors, plugins, agent products and service partners in one destination. Together, the catalog and submission portal serve both sides of a distribution market: developers need a route to publication and evidence of use, while customers need discoverable, reviewed tools. Anthropic says users can find more than 2,000 connectors and plugins in the broader marketplace. The number describes available integrations, not paid adoption or active usage. Its value will depend on whether the review process gives enterprises confidence without making the launch cycle too slow for partners. For Anthropic, each useful extension expands the set of tasks a customer can accomplish without leaving Claude.
OpenAI launched richer plugin surfaces and a marketplace days later, underscoring competition to become the default place where third-party AI workflows are discovered. Anthropic's path is grounded in MCP and skills, standards it has promoted for connecting tools and reusable instructions. That may appeal to developers who want components usable across environments, even as Claude-specific discovery remains valuable. The portal does not disclose a fee, revenue share or approval timetable for publishers. What it does deliver is a practical route from building a connector to measuring installed use. If enough developers find customers through the directory, Claude's utility can grow faster than Anthropic could expand its own applications team.
Analysis
The portal reduces the fixed cost for a developer to reach Claude users: one submission, an automated review and adoption metrics replace bespoke distribution for each enterprise. Anthropic earns indirectly if integrations make its usage-based model more central to customer workflows, while third parties retain the specialized application layer. The more than 2,000 available integrations provide catalog breadth, but the economic value is concentrated in those that lead to repeated, billable work. Safety review and managed authentication can make adoption easier for enterprises, though Anthropic bears operating costs in vetting the ecosystem. Competition with other assistant platforms turns developer attention into a scarce input; transparent analytics can help Anthropic win it.