- Review Vibe Coding and Vibe Learning
- Review Agency and Agentic
- Manually adding the Azure MCP Server to Cursor
- One-click installations of MCP servers from the Cursor Marketplace - Firecrawl and Hugging Face
- The Cursor Command-Line Interface (CLI)
- Skills
- A brief example of using Claude Code with portable Cursor skills
The intended purposes of this session is to:
- Increase your knowledge of MCP and Agentic AI
- Increase your productivity with Cursor to accelerate your learning
Remember my prediction from the first session?
Prediction: You will learn more from AI (i.e. - Cursor) than from Chris. I'm just your "travel guide".
Simple Vibe Coding - create some code.
Simple Vibe Learning - explain some code.
A review of this important topic from the previous session on MCP.
Agency is the sense of control that you feel in your life, your capacity to influence your own thoughts and behavior, and have faith in your ability to handle a wide range of tasks and situations
The key word here is control.
- You define the set of MCP Server(s) that the application can use
- The collective set of tools and resources (see above)
- Given a task to complete (i.e. - a prompt), the agentic application will operate autonomously
- It will determine what MCP Servers, tools, and resources to use, and in what sequence
- This is a NOT Developer-specified set of imperative logic
- Instead, the Agentic AI application figures out, itself, how to complete the task!
The key concept here is autonomous control.
Cursor can act as an Agentic Model Context Protocol (MCP) client using a set of custom and/or marketplace MCP servers. The MCP servers can be local to your computer or remote.
Installing Model Context Protocol (MCP) servers in Cursor can be done through a one-click installation or manual configuration using the mcp.json file. This process connects external tools and data sources to your AI agent within the Cursor
See the previous session-mcp-the-model-context-protocol session where we described the Model Context Protocol and created/deployed a local MCP server to Cursor. It uses my m26 python library for running calculations.
In this session we'll see more advanced uses of Cursor and MCP.
See https://learn.microsoft.com/en-us/azure/developer/azure-mcp-server/get-started/tools/cursor
Add the following configuration to the mcpServers mcp.json file:
"Azure MCP Server": {
"command": "npx",
"args": [
"-y",
"@azure/mcp@latest",
"server",
"start"
]
}
- macOS/Linux: ~/.cursor/mcp.json (the ~ stands for your home directory)
- Windows: %USERPROFILE%.cursor\mcp.json (also in your home directory)
After adding the Azure MCP Server the mcp.json file looks like the following. The m26-mcp-server was presented in the MCP session on 3/31.
{
"mcpServers": {
"m26-mcp-server": {
"command": "uv",
"args": [
"run",
"--project",
"/Users/cjoakim/github/zero-to-AI-private/python/m26-mcp-server.py",
"--with",
"fastmcp",
"fastmcp",
"run",
"/Users/cjoakim/github/zero-to-AI-private/python/m26-mcp-server.py"
],
"env": {},
"transport": "stdio"
}
},
"Azure MCP Server": {
"command": "npx",
"args": [
"-y",
"@azure/mcp@latest",
"server",
"start"
]
}
}
Npx (Node Package eXecute) is a powerful CLI tool bundled with npm (v5.2.0+) that allows you to run Node.js binaries and packages without permanently installing them globally. To install npx on Windows, you must install Node.js, as npx is automatically included
Install Node.js on Windows with the following command:
winget install OpenJS.NodeJS.LTS
Node.js is a high-performance version of JavaScript that runs outside of a web browser. JavaScript originally ran only within web browsers.
Rabbit hole: TypeScript is an excellent way to author JavaScript code, as it looks like C# and has a compiler. The compiled/transpiled code is solid and reliable.
Now we can ask the following in the Cursor Chat tab - "List my Azure storage accounts".
This MCP server connects to Azure using your az login credentials. Read more about the az program here.
Let's visit the Cursor Marketplace.
Marketplace home page looks like this.
Click into Firecrawl. It's a web scraping, crawling, and search for AI agents. Gives Cursor full access to web content through the Firecrawl CLI.
Simply click the blue Add to Cursor button.
In Cursor, navigate to the plugins configuration Cursor --> Settings --> Cursor Settings --> Plugins and see that Firecrawl is now installed.
See the saved firecrawl usage instructions as markdown.
Oh, surprise! Firecrawl requires a subscription! Regardless, this shows you how to install a MCP server or plugin from the Marketplace with one-click.
In Cursor, navigate to the plugins configuration UI Cursor --> Settings --> Cursor Settings --> Plugins.
Then enter Hugging Face in the search field. Hit enter.
Click the blue Add to Cursor button.
Ask Cursor how to use Hugging Face, and save the output as markdown.
See the saved hugging usage instructions as markdown.
Note: This section assumes that you have an account at Hugging Face, and have the hf CLI program installed (I do).
Now we can ask the following in the Cursor Chat tab - "list the top five trending hugging face datasets.".
Don't keep any unusable Plugins or MCP servers configured in Cursor. Otherwise, Cursor may try to use them.
For example, Cursor tried to use Firecrawl MCP server, even though I don't have a Firecrawl account. Therefore, I removed this plugin.
This is an alternative to the Cursor UI. You can use one, or both.
See https://cursor.com/docs/cli
[~/github/zero-to-AI]$ agent --help
Usage: agent [options] [command] [prompt...]
Start the Cursor Agent
Arguments:
prompt Initial prompt for the agent
Options:
-v, --version Output the version number
--api-key <key> API key for authentication (can also use CURSOR_API_KEY env var)
-H, --header <header> Add custom header to agent requests (format: 'Name: Value', can be used multiple times)
-p, --print Print responses to console (for scripts or non-interactive use). Has access to all tools,
including write and shell. (default: false)
--output-format <format> Output format (only works with --print): text | json | stream-json (default: "text")
--stream-partial-output Stream partial output as individual text deltas (only works with --print and stream-json
format) (default: false)
-c, --cloud Start in cloud mode (open composer picker on launch) (default: false)
--mode <mode> Start in the given execution mode. plan: read-only/planning (analyze, propose plans, no
edits). ask: Q&A style for explanations and questions (read-only). (choices: "plan",
"ask")
--plan Start in plan mode (shorthand for --mode=plan). Ignored if --cloud is passed. (default:
false)
--resume [chatId] Select a session to resume (default: false)
--continue Continue previous session (default: false)
--model <model> Model to use (e.g., gpt-5, sonnet-4, sonnet-4-thinking)
--list-models List available models and exit (default: false)
-f, --force Force allow commands unless explicitly denied (default: false)
--yolo Alias for --force (Run Everything) (default: false)
--sandbox <mode> Explicitly enable or disable sandbox mode (overrides config) (choices: "enabled",
"disabled")
--approve-mcps Automatically approve all MCP servers (default: false)
--trust Trust the current workspace without prompting (only works with --print/headless mode)
(default: false)
--workspace <path> Workspace directory to use (defaults to current working directory)
-w, --worktree [name] Start in an isolated git worktree at ~/.cursor/worktrees/<reponame>/<name>. If omitted, a
name is generated.
--worktree-base <branch> Branch or ref to base the new worktree on (default: current HEAD)
--skip-worktree-setup Skip running worktree setup scripts from .cursor/worktrees.json (default: false)
-h, --help Display help for command
Commands:
install-shell-integration Install shell integration to ~/.zshrc
uninstall-shell-integration Remove shell integration from ~/.zshrc
login Authenticate with Cursor. Set NO_OPEN_BROWSER to disable browser opening.
logout Sign out and clear stored authentication
mcp Manage MCP servers
status|whoami View authentication status
models List available models for this account
about Display version, system, and account information
update Update Cursor Agent to the latest version
create-chat Create a new empty chat and return its ID
generate-rule|rule Generate a new Cursor rule with interactive prompts
agent [prompt...] Start the Cursor Agent
ls Resume a chat session
resume Resume the latest chat session
help [command] Display help for command
Enter agent mode from the CLI.
See the three modes of Cursor.
- Agent mode - Full access to all tools for complex coding tasks
- Plan mode - Design your approach before coding with clarifying questions
- Ask mode - Read-only exploration without making changes
Let's have a conversation with the agent about the codebase.
Ask the agent a question.
Ouch, that unbiased perspective hurts! I wanted to appeal more to non-Developers.
Let's try again, ask it to summarize the series.
See the generated series summary output.
Note to self, lesson learned: For my next similar project, do the following:
- Articulate the goals of the series in a few bullet points in a markdown file
- As I develop the content, ask the agent to tell me if I'm on track in meeting the goals
- The unbiased (and sometimes brutally honest) feedback is invaluable
Agent is Cursor's assistant that can complete complex coding tasks independently, run terminal commands, and edit code. Access in sidepane with Cmd+I.
Agent Skills is an open standard for extending AI agents with specialized capabilities. Skills package domain-specific knowledge and workflows that agents can use to perform specific tasks.
Agent Skills are folders of instructions, scripts, and resources that agents can discover and use to do things more accurately and efficiently.
See file .agents/skills/roll-dice/SKILL.md in this repo.
Thus, the skills can be versioned in source-control (i.e. - GitHub).
The format of these skills is an Open Standard called Agent Skills IO.
Microsoft has created Azure Agent Skills - a set 193 of skills that are specific to Azure.
Claude Code is another AI tool like Cursor. It also conforms to and supports the Agent Skills Open Standard, thus your skills are portable across different AI tools and editors.
Claude Code is out-of-scope for this series, but here's a brief example of it.
Copied from the Cursor .agents directory to the Claude .claude directory.
├── .agents
│ └── skills
│ └── roll-dice
│ └── SKILL.md
├── .claude
│ ├── commands
│ │ └── application-review.md
│ ├── settings.json
│ └── skills
│ └── roll-dice
│ └── SKILL.md
And now the roll-dice skill works in Claude Code, too.
- Cursor Docs
- Cursor Docs MCP
- Cursor Docs Installing MCP Servers
- Cursor Marketplace
- Cursor Cookbook
- Cursor API
- Cursor CLI
- Cursor Skills
- Cursor Agent Best Practices
- Agent Skills IO
- Azure Agent Skills
- Azure MCP Server
- Hugging Face MCP Server