This chapter serves as a capstone project that weaves together the features learned from chapters 1 through 7.

It covers four expansion features: Skills, Schedules, MCP Connectors, and Plugins. When these four work in concert, Cowork transforms from a conversation partner into an autonomous digital colleague.

If you have grown accustomed to the method of issuing a single task and receiving results in prior chapters, this chapter builds a structure that eliminates the need for instructions themselves. Once designed, you get consistent results without repeating yourself each time, tasks execute automatically at set times, and data from multiple apps flows together in one unified stream.

However, the features in this chapter are areas where Cowork is evolving rapidly while in Research Preview, a testing phase before official release. The information presented here is based on details as of March 2026, but menu locations and specific options may change with updates. I recommend making a habit of checking the official Anthropic support page (support.claude.com) for the latest information.

1 Learning Your Own Standard Operating Procedures Through Skills

A Why You Need It: The Fatigue of Repeating the Same Instructions

Assume that every time you write a blog post for your company, you must follow these rules: keep the title to 40 characters or fewer, place the customer's pain point in the first paragraph, avoid directly naming competitors in the body, and include a "Get Started Now" button text at the end. What would happen if you had to deliver these rules to Cowork as a prompt each time? Some days you would forget to mention the competitor restriction, and other days you would word the call-to-action differently. The quality of the output would fluctuate based on how completely you remembered the prompt that day.

The same issue arises in code reviews. A senior developer has certain items they always check: whether error handling is missing, whether environment variables are hardcoded into the code, whether test coverage meets standards. Putting this checklist into a prompt every time is tedious, and leaving something out degrades review quality. Skills are a mechanism that puts an end to this repetitive explanation once and for all.

B What Is a Skill?

A Skill is a text file in Markdown (.md) format.

When you organize task procedures, tone and manner (the mood and speaking style of text), forbidden phrases, preferred formats, and reference examples into a single file and save it, Claude automatically reads that file and follows the guidelines whenever receiving a related request. You no longer have to give the same instructions every time.

Technically speaking, Skill files are stored in a local folder on your computer. Cowork usually reads only the summary information at the top of a Skill file, but loads and executes the entire file only when you issue a task that triggers it. Because of this "read only when needed" structure, performance remains fast even when you have dozens of Skills saved.

[Note] What is a Markdown file? Markdown (.md) is a lightweight format for writing formatted text. You can open it with a text editor, and it uses symbols to represent formatting like headings, lists, and emphasis. You can write it without any coding knowledge. You will use this format when editing Skill files directly.

C Creating a Skill: Basic Use

Follow along:

1. Open the Claude Desktop app and click the "Cowork" tab on the left side.

2. Start a new task and give an instruction as you normally would. For example, type the following:

"Write a blog post for our company. The topic is 'Inventory Management Tips for Small Business Owners'. Keep the title to 40 characters or fewer, and put the customer's frustration in the first paragraph. Don't mention competitor names directly. Unify the call-to-action text in the last paragraph to 'Start Free Today'."

3. Claude generates the output. If you are satisfied with it, type the following in the chat:

"Based on the rules I just gave you, create a 'Blog Writing Skill' that writes blog posts applying the same rules to any topic I give you in the future."

4. Claude automatically creates a Skill file (skill.md). The skill's name, objective, execution steps, and precautions are displayed on screen.

5. Click the "Copy to your skills" button. The Skill is saved to your local folder.

From now on, if you simply type "Write this week's blog post. The topic is 'How to Reduce Delivery Packaging Costs'" in a new conversation, Claude will automatically invoke the Blog Writing Skill and produce a post with all rules applied,title length, first paragraph structure, and call-to-action text.

D Using Skills: Practical Examples

Skills are not limited to writing. They apply to any recurring work with repeating rules.

Let's take email tone as an example. Gather 20 to 30 emails you have sent in the past into a single folder. Have Cowork open that folder, then instruct it like this:

"Analyze these emails, extract my unique tone of voice, greeting style, sentence length, and frequent expressions, then create 'My Email Tone Skill' that applies this style whenever drafting emails in the future."

Claude analyzes the patterns across these emails and organizes them into a Skill file. Later, when you simply request "Draft an email to vendor rep Kim about delivery schedule changes," you get an email with your natural tone, not the stiff style typical of AI.

You can also create an expense receipt processing Skill. If you save the rules for classifying receipt photos as a Skill (date format as YYYY-MM-DD, categories as meals/transport/office supplies/other, and flag items over 100,000 won), you can simply designate the receipt folder and say "Make the expense report" to get a completed Excel file.

E Using Skills: Real-World Application

The true power of Skills lies in converting organizational knowledge into files.

When expertise that existed only in a veteran employee's mind becomes a Skill file, the team's work quality remains stable even if that employee takes leave or departs. When a newly hired employee runs a task with that Skill applied in Cowork, they can produce results that meet team standards from day one.

When work methods change, simply edit the .md file. If brand color codes change, update that line. If new regulations add a legal review item, add one line to the Skill. You do not need to memorize a new prompt or reshare it with team members. Editing one file immediately applies the change to every task using that Skill.

Remember one principle when designing Skills: Claude does not infer implicit rules that are not written in the file. Instead of vague instructions like "be kind," write specific ones: "For non-refundable cases, first acknowledge the customer's concern, state the policy basis in one sentence, and offer an alternative in the final paragraph." Abstract Skills produce abstract results. Concrete Skills guarantee consistent quality.

F Troubleshooting

(1) If a Skill is saved but does not apply in new conversations: Confirm that the Skill file was saved in the correct location. Open the Skills list under Settings > Customize and check whether that Skill is enabled.

(2) If a Skill applies but some rules are ignored: The Skill file may be too long or rules may conflict with each other. Separate "must-follow rules" from "background information for reference," and place core rules at the beginning of the Skill file, keeping them short and clear.

(3) If you modify a Skill but the previous version continues to apply: Close the Claude Desktop app completely and reopen it. Changes to Skill files take effect when the app restarts.

2 24-Hour Background Automation Based on Schedules

A Why You Need It: Daily Preparation Tasks That Repeat Each Morning

Monday morning at 9 a.m. When you arrive at the office and turn on your computer, there is something you do first. You scan through unread emails and filter out the important ones, check today's and tomorrow's schedule, search for news in your area of interest, and identify items worth sharing with your team. This takes 30 minutes to an hour. It is not time spent analyzing content and making decisions, but time spent gathering the material needed to make those decisions.

Friday afternoon follows a similar pattern. You organize files piling up in your Downloads folder, transfer this week's expense receipts to the expense report spreadsheet, and prepare next week's meeting agenda. During busy or exhausting weeks, you skip this task, and a month later you find 200 files piled up in your Downloads folder.

The Schedule (Scheduled Tasks) feature automatically executes these repetitive tasks at a set time. Whether the user is sleeping or in a meeting, Claude wakes up on its own and handles the work.

What Is a Schedule?

A schedule is a reservation command that says "execute this task, at this time, in this cycle." You can set the desired cycle,hourly, daily, weekdays only, weekly, and so on,and specify the execution time precisely. When the scheduled time arrives, Cowork opens a work session on its own, accesses the designated folder, and performs the task using pre-saved skills and connectors.

There is one prerequisite. Schedule tasks work only when the user's computer is on and the Claude Desktop app is running in the background. If you shut down the app completely or the computer enters sleep mode, scheduled tasks will not execute.

Setting Up a Schedule: Basic Usage

Follow Along

1. Open the Cowork tab in the Claude Desktop app.

2. In the left sidebar, click "Scheduled."

3. Click the "New Task" button. Alternatively, you can type /schedule in an existing Cowork conversation window.

4. Enter the task content specifically. For example, write like this: "Create a file called Morning_Briefing.md in the 'Work' folder on my desktop every morning at 8 a.m. Structure the content like this. First, today's date and day of the week. Second, the titles and one-line summaries of three major AI-related news articles from web search. Third, today's to-do checklist template (five blank lines)."

5. Claude organizes and displays the task name, execution cycle, and task content. After confirming, click the "Schedule" button.

6. The newly created task is added to the Scheduled list in the left sidebar. If the computer is on and the app is running at 8 a.m. the next morning, the briefing file will be automatically generated in the desktop's 'Work' folder.

Using Schedules: Practical Examples

By changing the cycle and task content, various types of automation are possible.

Example of automatic file organization: "Scan the downloads folder every Friday at 5 p.m., move image files to the 'Photos' folder, PDFs to the 'Documents' folder, and the rest to the 'Other' folder. Save the list of moved files as 'Organization_Result_Date.txt' on the desktop."

Example of competitor monitoring: "Every Monday morning at 7 a.m., search the web for the past week's news about three companies,Coupang, Naver Shopping, and Kakao Commerce,and if there are any new service launches or policy changes, summarize them and save as a 'Competition_Weekly.md' file."

You can also combine skills created earlier with scheduled tasks. "When creating a summary of emails received every evening at 10 p.m., apply my Email Tone skill and attach a draft to emails that need a reply." Skills handle quality, and schedules handle timing.

Using Schedules: Real-World Application

There is something to remember about schedule automation. Auto-generated results are first drafts, not final reports.

News collected by AI in the early morning may contain errors and can miss context. In a competitor monitoring report, information stating "Company A launched a new service" might actually be from an article that is merely rumor. Automation prepares materials needed for decision-making in advance; it does not replace decision-making itself.

The most effective pattern in practice works like this. Schedule collects information in the early morning and does the first pass of organization. When the user arrives at work in the morning, they review the results in under ten minutes and adjust only the parts that need it. Preparation work that took one hour drops to ten minutes. Complete automation is not realistic; semi-automation that combines human review with AI preparation is the practical goal.

If You Encounter These Problems

(1) If a task does not run at the scheduled time: Check whether the Claude Desktop app is running and whether the computer has entered sleep mode. On Windows, adjust sleep settings at Settings > System > Power.

(2) If a scheduled task stops mid-execution: The network connection may have been lost, or you may not have access permission to the target folder. Click the task in the Scheduled list to view the execution log.

(3) If a "Usage limit reached" message appears: Schedule tasks are included in your general Cowork usage. Pro plan users may reach the limit quickly. You can set the execution model for scheduled tasks to

Sonnet 4.6 to save usage compared to Opus 4.6. Reducing the execution frequency from daily to weekdays is also an option.

3. Controlling External Apps Through MCP Connectors

Why Is It Needed: Time Spent Switching Between Tools

Consider a project manager preparing a weekly report. They check progress from each team in Notion, search for related documents in Google Drive, search for issues discussed this week in Slack, and compile all of this into a single report. Manually combining information by switching between four tools takes over one hour. That is time spent gathering information, not writing the report.

As business tools multiply, this time spent switching windows grows longer. Each tool has its own interface and search method, so data from the same project remains disconnected. You copy from app A and paste into app B dozens of times a day.

An MCP (Model Context Protocol,the specification for AI to exchange data with external apps or services) connector is a bridge that breaks through this disconnection. You can think of it as Claude's hand reaching out to other apps.

What Is an MCP Connector?

An MCP connector is a connection device that allows Claude to access external applications directly, read data, search, and in some cases write or edit content. Once a connector is linked, users need not open a separate browser tab; they can simply say "Get this week's meeting notes from Notion" within Cowork and receive the data.

As of March 2026, Anthropic officially supports roughly ten connectors including Google Drive, Gmail, Google Calendar, Slack, Notion, GitHub, Jira, Linear, Asana, DocuSign, and FactSet, with more being added. These official connectors connect instantly after a few clicks of authentication (OAuth, a process where users grant app access).

If you use a tool not on the official list, you can connect it using Cowork's custom connector feature if the service provides an MCP server address. Connecting Zapier MCP also makes it possible to access over 8,000 external apps.

First Connector Connection Tutorial: Basic Usage

For those setting up an MCP connector for the first time, I will guide you through the entire process of connecting a Google Drive connector.

Follow Along

1. In the Claude Desktop app, click the Settings icon at the bottom left.

2. Select the "Customize" menu and click the "Connectors" tab.

3. Find and click "Google Drive" in the connectors list.

4. Click the "Connect" button to open a browser window and display the Google account login screen. Log in with your Google account and allow Claude to access Drive.

⑤ Once authentication is complete, the connector status changes to 'Connected.'

⑥ Return to the Cowork tab and start a new task. Enter the following: 'Find the file titled 2024 Fourth Quarter Performance Report in Google Drive and summarize the key content in five lines.'

⑦ Claude accesses Google Drive, searches for the file, reads its content, and provides a summary.

D. Connector Use: Application Examples

When a single connector is linked, it functions as a convenient search tool. When multiple connectors are connected simultaneously, it becomes an automated workflow.

Let's revisit the weekly report preparation example mentioned earlier. With Notion, Slack, and Google Drive connectors all connected, you give this instruction:

'Collect progress updates from each team's Notion project board this week, summarize the issues discussed this week in Slack's #project-alpha channel, extract the list of related documents from Google Drive, and create a draft weekly report. Save the report as a Word file on the desktop.'

Data from three apps is collected in a single conversation, and an organized report is created. A task that took one hour now finishes with a single instruction.

With the Gmail connector linked, email tasks can also be handled within Cowork. When you request 'Find and summarize emails about contracts from Manager Kim last week,' Claude searches Gmail and organizes the essentials. This can extend to 'Draft your review comments and create a reply draft.' Reading the email, analyzing it, and writing a response connect as a single flow.

E. Connector Use: Practical Application and Security Considerations

The convenience of connectors comes with the weight of security. Since you are giving Claude access keys to company email, documents, and work channels, it is wise to establish some principles.

First, grant permissions narrowly. When only read access is needed, there is no reason to grant write access. Start connector settings as read-only, and add write permissions only when absolutely necessary.

Second, when handling sensitive data, first review organizational policy. Cowork is currently in research preview status, and audit logs or data export functions do not capture Cowork activity. Anthropic's official documentation also recommends against using Cowork for regulated workloads.

Third, do not forget that automation can make mistakes. It is possible for Claude to send messages incorrectly to Slack or enter wrong data into Notion. For important write operations, you must include a human approval step with an approach like 'Create a draft and show it to me; I will send it after reviewing.'

F. If These Problems Occur

(1) If you see 'Authentication failed' when connecting a connector: Check whether the account already logged in to the service (Google, Slack, etc.) in your browser is the one you want. Authenticating while logged in as a different account can connect the wrong account. You can avoid this problem by attempting authentication in an incognito mode browser.

(2) If the connector is connected but 'No results found' appears: The search scope might be too broad or the keyword might not be exact. Resubmit the request with a specific filename or keyword, such as 'Fourth Quarter Report.' With Google Drive, Shared Drive files may have different search scope settings, so verify that the scope is 'My Drive.'

(3) If external app responses are slow or timeouts occur: External services like Notion and Slack have rate limits. Requesting too much data at once can cause the service to refuse a response. Narrow the request scope to something like 'Last 7 days' and try again.

4. Implementing Department-Specific Custom Plugins

A. Why It Is Needed: The Effort of Setting Up Skills and Connectors Separately

You have learned how to create and configure skills, schedules, and connectors separately. A marketing team must set up content writing skills, brand guideline skills, Notion connectors, Slack connectors, and weekly posting calendar schedules separately. Someone familiar with IT can accomplish this, but having ten team members each replicate the same setup is a different scale of problem.

A plugin bundles these pieces into a single package. Skills, connectors, slash commands (shortcut commands starting with the / symbol), and sub-agents (auxiliary AI that shares complex tasks) are packaged together, so a single installation sets up the entire AI environment needed for a specific role.

B. What Is a Plugin

Plugins fall into two main categories.

One is the official plugin provided by Anthropic by default. On January 30, 2026, Anthropic Labs released 11 open-source plugins for Sales, Legal, Finance, Marketing, Data Analysis, Productivity, Product Management, and Software Development, among others. In February, 15 additional plugins for Design, Operations, HR, and others were added. These plugins can be viewed at claude.com/plugins and installed directly from Cowork.

The other is a custom plugin created by the user. After installing a standard plugin, you can click the 'Customize' button to modify it to suit your company, or you can build a new plugin from scratch.

C. Installing Plugins: Basic Usage

Follow along: ① Go to the Cowork tab in the Claude Desktop app.

② Click the 'Plugins' menu in the lower left.

③ A list of plugins appears. Select one from the plugins categorized by role, for example, 'Marketing.'

④ On the plugin detail screen, verify the list of included skills, slash commands, and connectors, then click the 'Install' button.

⑤ Once installation is complete, new commands appear when you type / in the Cowork conversation box. You can use commands like /brand-review, /campaign-plan, and /draft-content.

⑥ Execute the command. For example, when you enter /draft-content, Claude presents a form and asks 'Which channel is this for? (Blog/SNS/Email)' and 'What is the topic?' Fill in your answers, and content is generated with the skills built into the plugin applied.

D. Plugin Use: Application Examples and Customization

Standard plugins are designed for general use and may not fit your company's situation exactly. In that case, you customize.

Follow along: ① Click the 'Customize' button on the detail screen of the installed plugin.

② Claude asks questions. 'What CRM does your company use?' 'Who is your primary customer base?' 'What is your brand tone?' Answer these questions, and the skill files within the plugin are modified to suit your company.

③ The customized plugin can be shared with team members. For Enterprise plans, administrators can create a private plugin marketplace for organization-wide distribution.

Let me give an example of customizing a legal plugin. The standard legal plugin contains a general contract review checklist. To this, you add your company's risk tolerance criteria, frequently used contract clause templates, and a list of toxic clauses that must be removed. Then, place partnership contracts in a folder and run the /audit command, and it identifies risky clauses according to your company's standards and delivers a review result with risk levels color-coded.

[Note] What is a slash command? When you type a command after the / symbol in the Cowork conversation box, a specific task runs immediately. It is similar to app shortcuts on a smartphone. /schedule creates a scheduling task, and /brand-review starts a brand review. When you install a plugin, the slash commands included in that plugin are automatically added.

E. Plugin Use: Practical Application and Organization-Wide Expansion

The true value of a plugin appears at the organizational level, not the individual level.

Consider a structure where ten departments each have their own plugins, but all plugins share the same company brand skills. Whether marketing uses it, legal uses it, or HR uses it, the tone and format of documents remain consistent. Individual department-specific rules are handled by each department's plugin, while company-wide consistency is guaranteed by shared skills.

Let me give an example using a new employee's first day. A new hire joining the marketing team opens Claude Cowork and activates the marketing plugin. Without any separate training, running the /draft-content command produces content with the team's brand guidelines, tone and manner, and CTA rules all applied. Instead of making them read dozens of pages of manuals, a single plugin does that job.

Plugin updates are also managed centrally. When the law changes and you update the legal plugin's checklist, that change is immediately reflected to all employees using the legal plugin. You don't need to notify 100 people individually that 'from now on, do it this way.'

However, you should not accept plugin-generated results as final approval. Just because a legal plugin reviewed a contract doesn't mean you can skip the lawyer's judgment, and just because an HR plugin evaluated a resume doesn't mean it can replace an interviewer's eye. A plugin's role is to handle repetitive first-pass work; final judgment and decision-making belong to people. In a corporate environment, clarity on this boundary between AI adoption and human judgment is the difference between success and failure.

In the Enterprise plan, administrators can control which plugins are deployed to which departments, which connectors are allowed, and what data can be accessed. As of the February 24, 2026 update, a private plugin marketplace, per-user automatic installation, and admin dashboard features have been added.

If these issues arise

(1) If slash commands don't appear after installing a plugin: Exit the Claude Desktop app and restart it. Check that the plugin shows as "Active" in your plugin list.

(2) If your customizations don't apply to other team members: Personal customizations only apply to you. To roll out to the entire team, the Enterprise administrator can deploy through the plugin marketplace, or you can export the customized plugin as a ZIP file and have team members each upload it themselves.

(3) If plugin skills and personally-created skills conflict: When a plugin skill and a personal skill have different rules for the same type of task, Claude can become confused. Check the priority in Settings > Customize, and disable one of the overlapping skills.

The four functions covered in this chapter,skills, schedules, MCP connectors, and plugins,are useful independently, but their combined power shows the full potential of Claude Cowork. Skills manage quality, schedules manage time, connectors link tools, and plugins tie it all together. The journey that started in Chapter 1 with organizing a Downloads folder has now arrived at designing workflows for an entire organization.

You don't need to implement all features at once. Create one skill and use it for a week, set up one schedule and check the results each morning, connect one connector and experience the convenience of search, and then naturally move to bundling with plugins. Just start with what you need right now.