# How to Build Multi-Step AI Agents with CLAUDE’s mCP Update

- Link: https://mohammadjazim.blog/2024/11/30/how-to-build-multi-step-ai-agents-with-claudes-mcp-update/
- Published: 2024-11-30T00:56:14+05:30
- Author: Mohammad Jazim

The latest update to Claude, the **Model Context Protocol (mCP)**, is a game-changer
for AI enthusiasts. It turns Claude into an interactive API that can communicate
with external tools and apps, enabling you to build complex, multi-step AI agents
that can execute tasks autonomously—all with a single prompt.

Here’s how you can set it up and use it to automate complex workflows.

For a practical example of connecting research capabilities to agent workflows, 
see my walkthrough of [building research agents with Exa Agent and Exa MCP](https://mohammadjazim.blog/2026/06/07/building-ai-research-agents-exa-agent-mcp/).

---

### What Is mCP?

**Model Context Protocol (mCP)** is a new feature that lets Claude interact with
external tools, servers, and apps.

This addresses a common issue in building AI agents: connecting them to your data
and enabling them to act on it. With mCP, Claude becomes more than a chatbot—it 
evolves into an **AI agent** capable of running workflows, automating tasks, and
even handling coding operations.

Imagine having an AI agent that can browse the web, manage GitHub repositories, 
and perform multi-step actions like a full-stack developer—all from a single prompt.

Let’s dive into the steps to set this up.

---

### Step 1: Download CLAUDE Desktop

 1. **Download the app**: Visit Claude’s download page and choose the version for your
    operating system (MacOS or Windows).
 2. **Install it**: Once downloaded, move the app into your Applications folder and
    open it.
 3. **Log in**: If you don’t have a Claude account, create one—it’s worth it.

---

### Step 2: Set Up the mCP Configuration

Now we’ll configure Claude to interact with external tools.

**Open the Terminal**: Use the following command to open Claude’s configuration 
folder:

> open ~/.claude-desktop

**Create a Configuration File**: Run this command to create a config.json file in
the folder:

> touch ~/.claude-desktop/config.json

**Edit the Configuration File**: Open the file using a text editor (e.g., VS Code,
Cursor, or Notepad). Add the following JSON code to define your tools:

{

“servers”: {

“brave_web_search”: {

“type”: “search”,

“api_key”: “YOUR_BRAVE_API_KEY”

},

“github”: {

“type”: “code”,

“personal_token”: “YOUR_GITHUB_TOKEN”

}

}

}

**Save the file** and restart the Claude app to load your new configuration.

---

### Step 3: Set Up API Keys

To connect Claude to external tools, you’ll need API keys:

 1. **Brave Search API**:
 2. **GitHub Personal Token**:

---

### Step 4: Launch Claude with mCP Tools

Restart Claude, and you’ll see the mCP tools listed under the app’s interface. Now,
Claude can:

 * Browse the web using Brave Search.
 * Interact with GitHub to create repositories, push code, and handle pull requests.

---

### Step 5: Automate Multi-Step Tasks

Here’s where it gets exciting. With everything set up, you can automate complex 
workflows. For example:

### Prompt:

    ```wp-block-preformatted
    Please do the following: 
    1. Create a new HTML file for a webpage. 
    2. Push it to a GitHub repository called "new-project". 
    3. Add CSS styling to the page and push the update. 
    4. Create a GitHub issue suggesting enhancements to the webpage. 
    5. Create a feature branch, make the changes, and submit a pull request.
    ```

### What Claude Will Do:

**Create a Repository**:

> git init new-project

> git remote add origin [https://github.com/YOUR-USERNAME/new-project.git](https://github.com/YOUR-USERNAME/new-project.git)

> git push -u origin main

**Generate Files**: Claude writes the following index.html and style.css files:

<!– index.html –>

<!DOCTYPE html>

<html lang=”en”>

<head>

<link rel=”stylesheet” href=”style.css”>

<title>New Project</title>

</head>

<body>

<h1>Welcome to New Project</h1>

</body>

</html>

/* style.css */

body {

font-family: Arial, sans-serif;

background-color: #f4f4f4;

text-align: center;

padding: 20px;

}

**Push Files to GitHub**:

> git add .

> git commit -m “Initial commit: Add HTML and CSS”

> git push origin main

**Create an Issue**: Claude uses the GitHub API to create an issue titled “Enhance
the homepage”.

**Create a Feature Branch**:

> git checkout -b feature/homepage-enhancements

**Submit a Pull Request**: Claude creates a pull request from the feature branch
to the main branch.

---

### Step 6: Test Your AI Agent

With your setup complete, you now have a multi-step AI agent powered by Claude and
mCP tools. Test it out by providing different prompts and watching it execute tasks
autonomously.

---

### The Future of AI Agents

This setup demonstrates how accessible and powerful AI agents have become. With 
tools like Claude and mCP, anyone can build AI-powered workflows that save time,
automate repetitive tasks, and unlock new levels of productivity.

Whether you’re managing code, browsing the web, or running a business, the future
is here—and it’s powered by AI agents.

### Share this on:

 * [  LinkedIn ](https://mohammadjazim.blog/2024/11/30/how-to-build-multi-step-ai-agents-with-claudes-mcp-update/?share=linkedin)
 * [  WhatsApp ](https://mohammadjazim.blog/2024/11/30/how-to-build-multi-step-ai-agents-with-claudes-mcp-update/?share=jetpack-whatsapp)
 * [  Email ](https://mohammadjazim.blog/2024/11/30/how-to-build-multi-step-ai-agents-with-claudes-mcp-update/?subject=%5BShared%20Post%5D%20How%20to%20Build%20Multi-Step%20AI%20Agents%20with%20CLAUDE%E2%80%99s%20mCP%20Update&body=https%3A%2F%2Fmohammadjazim.blog%2F2024%2F11%2F30%2Fhow-to-build-multi-step-ai-agents-with-claudes-mcp-update%2F&share=email)
 * [  X ](https://mohammadjazim.blog/2024/11/30/how-to-build-multi-step-ai-agents-with-claudes-mcp-update/?share=twitter)
 * [  Facebook ](https://mohammadjazim.blog/2024/11/30/how-to-build-multi-step-ai-agents-with-claudes-mcp-update/?share=facebook)

### Like this:

Like Loading…
