Security & AI news

AI agent ecosystems · By FireAI Security & Research Team · Published

Anthropic opens Claude Marketplace with more than 2,000 plugins and connectors

Anthropic launched Claude Marketplace with over 2,000 connectors and plugins from Atlassian, Google, Microsoft and others; BleepingComputer’s report describes no security requirements.

A library of shelves and the FireAI mascot holding a block sign, next to the words “Claude Marketplace opens with 2,000+ plugins.”

Anthropic has launched Claude Marketplace, a catalogue that brings plugins, connectors, agents and related tools into one place, BleepingComputer reported on 27 September 2026. The catalogue offers more than 2,000 connectors and plugins at launch [1].

Background

A connector lets an AI assistant read from or act in another service, for example a document store or an issue tracker. Developers build them with the Model Context Protocol (MCP) and Agent Skills, according to the report. Each connector widens what the assistant can reach, and each one is code or a server run by a party other than the user. Recent FireAI News items describe how such components have been abused: an MCP server that waited before hunting for credentials and a plugin flaw in AI coding agents.

What the report describes

Plugin and connector providers named include Atlassian, Google, Microsoft, Notion and Salesforce. Partners offering Claude-powered agents or products include CrowdStrike, Cursor, Harvey, Legora, Lovable and Snowflake, and consulting partners include Accenture, Boston Consulting Group and Deloitte. Users can reach pre-integrated tools, partner agents and consulting help for deployment in an organisation, and companies can apply to list Claude-powered products [1].

BleepingComputer notes that Anthropic wants to avoid the fate of OpenAI's earlier app marketplace, which the report describes as unsuccessful, by allowing broader developer participation, and it says only time will show whether the approach works [1]. The report as reviewed does not describe permission models, review procedures or security requirements for listings.

Implications for Mac users

Users of Claude, and of other assistants with similar catalogues, gain convenience and a larger set of third-party components to trust. A catalogue of this size also makes it harder for any one person to know who wrote each component, who operates the server behind it and what it may do with the data it receives. The report gives no information about how listings are checked, so any statement about the safety of the catalogue would be a guess [1]. The relevant question for a Mac user is which connectors run locally and which reach remote servers.

Recommendations

  1. Enable only the connectors you use, and prefer those from a vendor you already trust.
  2. Read the permissions a connector asks for, and decline write or delete access when reading is enough.
  3. Remove connectors that you no longer use, and review the list after each assistant update.
  4. Keep sensitive accounts out of the assistant unless the task requires them.

Relevance to FireAI

FireAI does not review plugins or judge what a connector does with data after the assistant hands it over. On the Mac it shows the connections that local components make in Activity, lets per-app rules limit them and shows destinations on the world map. It cannot see what a hosted connector does on a remote server.

Limitations

This item relies on one report of the launch. It does not state pricing, availability by region or how listings are reviewed, and those may have been published by Anthropic elsewhere.

Try FireAI, by HisnLabs free for 17 days.

Sources

  1. BleepingComputer (Mayank Parmar), 27 September 2026: Anthropic turns Claude into an AI marketplace with 2,000+ plugins and connectors