Artificial intelligence
In early 2026, Anthropic introduced an ecosystem of plugins for Claude Cowork, its enterprise assistant. These plugins automate specialized workflows (marketing, sales, customer support, finance, research), extending Claude’s capabilities beyond conversation. Anthropic offers 11 open source plugins (marketing operations, information retrieval, sales, support, finance, legal, product management, biological research…) and enables companies to create and share their own plugins.
Modular architecture: each plugin is made up of four elements – “skills” (predefined skills), connectors (APIs and databases), slash commands and sub-agents. This file-based architecture facilitates creation and maintenance.
Customization and sharing: companies can modify existing plugins or create new ones. The marketplace allows for internal sharing within the organization, so that all teams have access to the same tools.
Cross-functional automation: plugins cover a wide range of areas (marketing, sales, finance, support, legal). They enable Claude to carry out actions such as generating documents, querying databases, triggering workflows, analyzing contracts or making forecasts.
Feedback and improvement: users can specify how Claude should manage a workflow and select the tools or data needed. Plugins improve thanks to human feedback.
Unlike other wizards, Cowork relies on an ecosystem of 11 open source plugins covering the entire operational spectrum of a company: Productivity, Enterprise Search, Marketing, Sales, Data Analysis, Customer Support, Plugin Creation/Customization, Finance, Legal, Product Management and Biology Research. Each plugin contains a set of dedicated skills and sub-agents; for example, the Marketing plugin generates content and plans campaigns, while the Customer Support plugin summarizes tickets and proposes solutions using knowledge bases. The composability of these file-based elements allows developers to add or remove functionality as required.
Anthropic provides a plugin creator that simplifies the creation of new plugins. Users can describe the workflow in natural language, select connectors (CRM, internal database, Web API) and define slash commands to execute specific actions. The system will automatically generate the file structure (skills, connectors, sub-agents), which can then be modified and enhanced. Workflows can include conditional decisions and calls to external services. For example, a marketing team might develop a plugin that retrieves CRM data, segments customers, generates personalized emails and schedules delivery according to budget. A legal department can create a plugin to analyze NDAs, extract contentious clauses and send a summary to the lawyer in charge.
Users can finely control the data and tools available for each plugin. One article points out that Claude can be told how to process a workflow, which connectors to use and what personal information is allowed. This flexibility makes it possible to adapt agents to internal policies and easily integrate proprietary data sources.
The 11 plugins cover a wide range of fields and can be combined to create complex solutions. Here are a few examples to illustrate the possibilities offered by this architecture:
Productivity: this plugin handles cross-functional tasks such as scheduling meetings, writing minutes and generating scripts. It integrates with calendars (Google Calendar, Outlook) and note services to automatically organize users’ days. Teams can thus delegate the task of taking notes and summarizing decisions made during meetings.
Enterprise Search: enables semantic searches of an organization’s knowledge bases and intranets. For example, an analyst might ask Claude to retrieve previous reports or compare offers, and the plugin will leverage a document index to return the most relevant information.
Marketing: this plugin generates multi-channel campaigns, adapts the tone according to customer segments and proposes budgets. It can analyze past performance, recommend optimizations and create content (emails, social network posts) aligned with brand guidelines.
Sales: by integrating with CRMs, this module qualifies leads, generates sales proposals and schedules follow-ups. Claude can automate the sending of follow-up e-mails, or suggest discounts and complementary products based on purchasing behavior.
Data analysis: finance or data teams can run queries on data lakes, generate graphs and detect anomalies without writing code. This plugin simplifies the preparation and exploration of large datasets, and can be coupled with predictive models.
Customer support: sorts incoming tickets, provides pre-written answers and escalates complex cases to human resources. This module can connect to incident management systems (Jira Service Management, Zendesk) and the internal knowledge base to provide rapid solutions.
Finance: automate invoice processing, generate accounting reports and monitor budgets. The plugin applies internal rules to detect unusual expenses and suggest corrective actions, helping finance teams to save time.
Legal: This module supports legal professionals in drafting contracts, extracting sensitive clauses and validating confidentiality conditions. It can analyze large volumes of legal documents to extract key points and draft summaries.
Product management: helps plan roadmaps, define requirements and prioritize functionalities by combining customer feedback and technical constraints. Product managers can generate user stories and monitor development progress.
Biology research: this plugin provides tools for analyzing genetic sequences, consulting scientific publications and generating experimental hypotheses. Researchers can automate literature searches and article synthesis.
These examples show that the plug-in ecosystem supports all key business functions. Plugins can be combined to create cross-functional workflows, for example orchestrating research, data analysis and customer support at the launch of a new product. Administrators can define governance rules for each plugin, restrict access to certain sensitive data and track actions via audit logs. As the marketplace grows, new plugins will enrich the ecosystem, creating a virtuous circle of innovation within companies.
According to a survey cited by Anthropic, 79% of companies are experimenting with AI agents, but only 8.6% have them in production, due to the complexity of integration and governance issues. Cowork plugins aim to bridge this gap by providing ready-to-use bricks and an open source extension model. Anthropic has also introduced persistent memory for Claude and integrations with Google Workspace and Asana, enabling plugins to access shared documents and track cross-functional tasks. These enhancements are part of Anthropic’s strategy to make Cowork a complete ecosystem, competing with suites from Google and Microsoft.
The open source model encourages community participation: external developers can publish new plugins or contribute to existing ones. Anthropic also plans to open an internal marketplace for companies, so that they can share plugins across the organization. For the moment, plugins are registered locally and sharing is done manually, but the company is announcing a future secure distribution mechanism. This transparency is intended to foster trust and adoption by large organizations, particularly those that need to validate the origin and security of components.
Extreme flexibility: the file-based, open-source structure enables plug-ins to be created and modified rapidly. Companies can adapt workflows to their needs and share internal solutions.
Business-oriented: the 11 plug-ins cover major operational functions. As a result, companies can get started quickly on concrete use cases.
Security and governance: although the platform is open, Anthropic provides granular access control and usage monitoring to ensure compliance.
Dependence on the Claude ecosystem: plugins only work with the Claude Cowork assistant. Organizations using other platforms must create specific connectors.
Support still under research: the plug-in program is launched in a “research preview” version and may be subject to change. Some advanced features (generalized organizational sharing) are not yet available.
Marketplace competition: other players (Salesforce, Google, Amazon) offer similar extension stores. Anthropic will have to prove the robustness and security of its plugins to convince companies.
| Solution | Strengths | Weaknesses |
|---|---|---|
| OpenAI Frontier | Comprehensive agent platform with identity management and context sharing | Opaque pricing and limited availability |
| Salesforce Agentforce | Agents that execute actions according to business rules, integrated with CRM and Service Cloud | Requires Salesforce environment and good data preparation |
| H2O.ai h2oGPTe | AutoML and GenAI platform with RAG integration and fine-grained access policy | Less focused on conversational plug-ins and more on model generation |
| AWS Bedrock + AgentCore | Agents and models via AWS with enhanced governance and persistent memory | Requires AWS environment and heavier technical configuration |
What are Anthropic Cowork plugins?
– Customizable extensions that add skills, connectors and sub-agents to the Claude Cowork assistant to automate specialized workflows (marketing, support, finance, etc.).
How do I create a plugin?
– By combining files describing skills, connectors and sub-agents, companies can edit open source templates and share their creations.
Main benefits?
– Flexibility, broad business coverage, customizability and shareability. Plugins let you adapt Claude to specific processes.
Limits?
– Necessity of using Claude, preview features, and competition with other marketplaces.