Controlling My Mac With Claude Code + "LLM OS" MCP Servers
Based on All About AI's video on YouTube. If you like this content, support the original creators by watching, liking and subscribing to their content.
Claude Code can orchestrate real Mac actions by connecting to multiple MCP servers, turning natural-language commands into desktop automation.
Briefing
A Mac can be turned into an “LLM OS” control surface by wiring multiple MCP servers into Claude Code, letting an AI trigger real desktop actions—open browsers, read and write the clipboard, gather system telemetry, and even run multi-step jobs that finish by putting the computer to sleep. The most striking demonstration is an autonomous workflow that searches for the latest AI news, compiles it into an HTML newsletter, emails it, and then shuts the Mac down—completed in roughly two minutes.
After connecting to nine MCP servers (including a global Gemini server), the setup supports quick, command-like actions. One example opens a curated set of Chrome tabs—LinkedIn, Reddit, Hacker News, AI Studio, ChatGPT, email, YouTube—using a single instruction. The same browser control can be extended to targeted navigation, such as jumping to Hacker News and then filtering to AI-related items before opening a specific story.
The workflow automation goes beyond browsing. A “job” prompt instructs the system to find the latest AI news from the past week, compile the results into a formatted HTML newsletter, and send it to the user’s email. The process uses tools like Brave search for discovery and fetch for pulling content from discovered URLs. Once the email is sent, the Mac is allowed to sleep; when the machine wakes again, the task is already marked complete. The delivered newsletter (“Latest AI News Weekly Update May 2025”) includes multiple top AI stories and a clean, readable layout, with the entire run taking about 120 seconds.
Clipboard control is another centerpiece. The MCP “clipboard” tool can pull whatever is currently copied in Chrome/elsewhere into the Claude Code context for actions like summarizing or explaining. The reverse direction also works: terminal commands can take text produced by the MCP/LLM workflow and write it back to the clipboard for pasting into other apps.
Several utility MCP servers round out the system. An app launcher tool opens specific development or recording profiles—such as launching Cursor alongside preselected URLs (AI Studio, ChatGPT, and a VT-related page). A Finder-based file tool locates files and opens them in Finder. A system MCP tool provides hardware and OS details (MacBook Pro with M3 Pro, 12 cores, 36 GB memory), network checks (pinging Google.com with packet loss and latency), process listing, and thermal/battery summaries (including CPU usage, memory usage, display sleep timing, and battery percentage).
Finally, the transcript shows a combined, sequential workflow: collect hardware and OS info, use it to search online for comparisons against newer Mac models (specifically M4 vs M3), generate a report on differences and upgrade considerations, and email the result. The comparison concludes that the M4 brings notable improvements, while the M3 remains capable for many professional workloads like heavy AI/ML and video editing. Overall, the demonstrations position MCP + Claude Code as a practical way to orchestrate everyday desktop tasks and longer multi-step automations through natural-language commands.
Cornell Notes
Claude Code paired with multiple MCP servers turns a Mac into an AI-controlled workstation. The setup connects to nine MCP servers and enables actions like opening curated Chrome tabs, filtering and opening Hacker News items, and launching app profiles (e.g., Cursor with preselected URLs). A standout automation searches Brave for the latest AI news, fetches the relevant pages, compiles an HTML newsletter, emails it, and then puts the Mac to sleep—finishing in about 120 seconds. Clipboard MCP support lets text flow both directions: copy from the desktop into the AI for summarizing/explaining, or write AI output back into the clipboard for pasting. System MCP tools provide hardware, network, and thermal/battery telemetry, which can be combined with web research to generate and email upgrade comparison reports.
How does the system control Chrome in a way that feels like “desktop automation,” not just chat?
What makes the newsletter workflow more impressive than basic browsing?
How does clipboard MCP change what an AI can do with everyday text?
What kinds of system-level capabilities are exposed through MCP here?
How does the setup handle a longer “sequential order” workflow that mixes local data and web research?
Review Questions
- What are the key steps in the autonomous newsletter workflow, and which MCP tools are used for discovery and content retrieval?
- Describe two ways clipboard MCP can be used in practice, including how text moves between the desktop and Claude Code.
- Which system MCP capabilities are demonstrated (hardware, network, thermals/battery), and how are they used in the upgrade-comparison workflow?
Key Points
- 1
Claude Code can orchestrate real Mac actions by connecting to multiple MCP servers, turning natural-language commands into desktop automation.
- 2
A curated Chrome MCP workflow can open multiple predefined tabs and then support follow-up navigation like filtering Hacker News to AI topics.
- 3
An autonomous job can search the web, fetch sources, generate an HTML newsletter, email it, and then put the Mac to sleep—reportedly completing in about 120 seconds.
- 4
Clipboard MCP enables two-way text flow: pull copied content into the AI for summarization/explanation and push AI output back to the clipboard for pasting.
- 5
App launcher MCP tools can open specific development or recording profiles, including launching Cursor alongside relevant URLs.
- 6
System MCP tools expose hardware, OS, network diagnostics, and thermal/battery telemetry that can feed into higher-level reports.
- 7
Combining local system facts with web research can produce an emailed upgrade comparison (M4 vs M3) with practical guidance for workloads like AI/ML and video editing.