MCP Integration: How Brave Search and Claude Desktop Enhance AI Agentic Assistant Capabilities

Introduction to MCP Agentic AI

The Model Context Protocol (MCP) has revolutionized how AI assistants interact with external data sources, offering smooth integration with tools, repositories, and local or cloud-based datasets. Introduced by Anthropic in late 2024, MCP enables AI to go beyond its traditional constraints, making it more proactive, contextual, and integrated into our workflows. This article focuses on setting up the Brave Search MCP plugin for Claude Desktop to strengthen your AI assistant with advanced web search capabilities. Whether you are a developer or a casual user, this guide will help you integrate this tool to use AI’s full potential. This is a continuation of the article that Rick recently wrote on Setting up Claude Filesystem MCP, but it is a standalone article. First, we’ll explore an in-depth discussion of MCP, followed by practical hands-on use case that show the Brave search connector with the Claude client. This hands-on approach will help you understand the power of the MCP architecture.

This hands-on guide will walk you through setting up Brave Search integration with Claude Desktop using the Model Context Protocol (MCP). Beyond adding search features, you’ll learn about MCP’s potential to transform how AI assistants connect with external tools. The guide covers API key setup, Claude Desktop configuration, integration testing, and troubleshooting. You’ll gain practical experience with MCP while enhancing Claude’s search capabilities.

Learning MCP hands-on is valuable for developers and technologists who want to explore AI agent development. You can apply these principles to build your own MCP integrations that connect AI assistants with databases, APIs, and custom tools. Whether you want to enhance Claude Desktop today or explore the future of AI integration, this guide offers a practical foundation for working with this powerful protocol.

Setting up Brave Search Agentic MCP plugin for Claude Desktop

Introduction: Unlocking the Power of MCP with Agentic Client-Server Architecture

The Model Context Protocol (MCP) represents a major advancement in the integration of artificial intelligence (AI) with external data sources. By creating a standardized framework for connecting AI systems to tools, repositories, and datasets, MCP bridges the gap between isolated AI assistants and the information they need. This helps deliver contextually aware and capable results. Central to its design is a client-server architecture that provides secure data access.


MCP Hosts: Empowering Agentic Applications

MCP Agentic Host secures access to data and user resources

At the core of the MCP framework areMCP hosts, which act as the central hubs for applications that require data access. Applications like Claude Desktop use these hosts to orchestrate and manage multiple connections to various data sources. The host ensures security, enforces permissions, and aggregates the data needed to enhance the AI’s contextual understanding, making tools like Claude more responsive, informed, and powerful.

### MCP Clients: The Agentic Communication Bridge

The clients in MCP architecture serve as the intermediaries between the host application and the data servers. Each client establishes a stateful session with a specific MCP server, ensuring a focused and secure communication channel. These clients handle requests for data or actions, route messages between the host and the server, and ensure a clear delineation of responsibilities, maintaining isolation between different data sources.


Agentic Communication Bridge

### MCP Servers: Enabling Access to Agentic Functionality

At the other end of the architecture are the MCP servers, which expose the functions of various data sources. Whether it’s accessing files on a local filesystem, retrieving records from a database, or searching the web via Brave Search, MCP servers provide the tools and interfaces necessary for these operations. Each server is specialized, focusing on a specific type of resource or function, ensuring modularity and composability within the MCP ecosystem.


MCP Servers: Serving up Agentic Resources

### Local and Remote Resources: The Agentic Data Backbone


Providing Access to Agentic Tools: The Agentic Data Backbone

The true power of MCP lies in its ability to connect AI assistants with local and remote resources, which serve as the actual repositories of data. These resources can range from local files on a user’s computer to remote APIs, databases, or cloud-based services. By integrating these resources into the AI workflow, MCP allows AI systems to access, analyze, and act upon the data required for many applications.

Bridging AI and Agentic Data: The MCP Promise

Together, these components form a cohesive architecture that transforms AI assistants from isolated tools into dynamic agents capable of accessing and utilizing a wide array of data sources. By combining host applications, communication clients, specialized servers, and diverse resources, MCP simplifies integration and expands the horizons of what AI can achieve. With this architecture, users can unlock a future where AI works more intelligently.


Providing Servers providing Agentic Resources, Clients communicating with Servers and Host Applications initiating the process and protecting access to the resources.

What is Brave Search MCP?

Brave Search MCP acts as a bridge between Claude Desktop and Brave’s privacy-focused web search engine. This integration allows Claude to perform both web and local searches easily, bringing the power of MCP’s structured connectivity to your fingertips.

With Brave Search MCP, Claude can:

• Retrieve real-time information from the web.
• Perform local context-based searches for enhanced results.
• Enable privacy-respecting AI-assisted research.


Brave Search MCP plugin provides Agentic Access to Web Search

Now that we have covered the background information, let’s set up Brave Search with the Claude client. This hands-on experience will help you understand the power of MCP once you see it integrate with one of the plugins.

Prerequisites

Before diving into the setup, ensure you have:

-Claude Desktop: Installed on your system (Download Claude Desktop). -Node.js: Installed and verified (node --version). -Brave Search API Key: Acquired from the Brave Search Developer Dashboard. -Terminal and Text Editor: Tools like VS Code or any preferred editor.

Setting Up Brave Search MCP Agentic Plugin for Claude Desktop

Step 1: Obtain the Brave Search API Key

1. Visit the Brave Search API.
2. Sign up or log in.
3. Select a plan (free tier supports 2,000 queries per month).
4. Generate your API key and store it securely.

Step 2: Configure Claude Desktop

Locate Configuration File

The Claude Desktop configuration file resides at:

~/Library/Application Support/Claude/claude_desktop_config.json

Add Brave Search MCP Server Configuration

Open the file in your editor:

code ~/Library/Application\ Support/Claude/claude_desktop_config.json

Insert or merge the following configuration:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "/Users/richardhightower/src",
        "/Users/richardhightower/clients"
      ]
    },
    "brave-search": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-brave-search"
      ],
      "env": {
        "BRAVE_API_KEY": "YOUR_API_KEY_HERE"
      }
    }
  }
}

This configuration step is key for enabling the Brave Search MCP integration with Claude Desktop. The JSON configuration file specifies how Claude Desktop should interact with the Brave Search service. It includes the needed command-line arguments and environment variables.

After adding or modifying this configuration, you must completely restart Claude Desktop for the changes to take effect. Closing and reopening the application window is not enough - you must fully quit Claude Desktop (using ⌘+Q on macOS or the equivalent on your operating system) and then relaunch it.

The restart allows Claude Desktop to initialize the MCP servers with the new configuration. It establishes the connections for the Brave Search integration to function.

Remember, replace YOUR_API_KEY_HERE with the API key obtained in Step 1.

If starting from scratch, use:

cat > ~/Library/Application\ Support/Claude/claude_desktop_config.json << 'EOL'
{
  "mcpServers": {
    "brave-search": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-brave-search"
      ],
      "env": {
        "BRAVE_API_KEY": "YOUR_API_KEY_HERE"
      }
    }
  }
}
EOL

The alternate approach shown above uses the Unix/Linux cat command with a “here document” (heredoc) to create the configuration file in one step. This method is useful when setting up the configuration for the first time. It creates the entire file structure at once without needing to manually edit an existing file.

The heredoc syntax (<< 'EOL') allows you to input multiple lines of text until the delimiter (EOL) is reached. The single quotes around EOL prevent variable expansion. This ensures the configuration is written exactly as specified.

This command will create the configuration file in the correct location with the proper JSON structure. Remember to replace YOUR_API_KEY_HERE with your actual Brave Search API key before running the command.

Step 3: Restart Claude Desktop

After saving changes:

1. Quit Claude Desktop (⌘+Q on macOS).
2. Relaunch the application to load the new configuration.

Claude Desktop


Open Claude Desktop and click the tools icon (hammer symbol) in the bottom right corner of the initial chat window. This will show you that the Brave Web Search Plugin is installed.

You should see the following tools for Brave Search.

brave_local_searchSearches for local businesses and places using Brave’s Local Search API. Ideal for queries about physical locations, businesses, restaurants, and services. Returns comprehensive details including business names, addresses, ratings, review counts, phone numbers, and opening hours. Use this for searches that imply “near me” or mention specific locations. If no local results are found, it automatically switches to web search.

From server: brave-search

brave_web_searchPerforms a web search using the Brave Search API, ideal for general queries, news, articles, and online content. Use this for broad information gathering, recent events, or when you need diverse web sources. Supports pagination, content filtering, and freshness controls. Maximum 20 results per request, with offset for pagination.

From server: brave-search

Testing the Integration

Basic Commands

1.Web Search:

```
Can you search for recent trends in generative AI?

```

2.Local Search:

```
Find restaurants near [your location].

```

Verification

Observe whether Claude retrieves relevant results. If issues arise, refer to the troubleshooting section.

Don’t worry, I will walk you through a couple of sample sessions with the Claude Desktop and Brave search.

Troubleshooting Common Issues

Configuration Not Loading

-Verify JSON Syntax: Use a JSON validator to check for errors. -Permissions: Ensure proper read/write permissions for the configuration file. -Correct Path: Double-check the file’s location.

API Key Problems

• Ensure the API key is accurately copied without extra spaces.
• Verify the key’s status in the Brave Search dashboard.

Node.js Issues

• Check installation: node --version.
• Reinstall Node.js if necessary.

Enhancing Security

To improve security and protect sensitive data:

• Store API keys securely in environment variables or encrypted vaults.
• Rotate API keys regularly to reduce security risks.
• Restrict configuration file access to trusted users only.

Sample Session with ClaudeMe:

Can you search for recent news about artificial intelligence?Claude:

I’ll search for recent news about AI using the Brave search function.

Let me search for some more specific recent developments.

Based on these search results, I can provide an overview of recent AI developments and regulatory trends:

1. Regulatory Landscape: 2024 has seen significant movement in AI regulation:
   • At least 40 U.S. states, along with Puerto Rico, the U.S. Virgin Islands, and D.C., have introduced AI-related legislation
   • The European Union has established comprehensive rules through the AI Act and AI Liability Directive
   • The African Union is developing a continent-wide AI strategy to protect consumers and help African countries compete in the AI space
2. Scientific and Technical Developments:
   • MIT researchers are working on new AI methods that find novel connections between science and art to discover new materials
   • In the medical field, there have been advances in specialized AI applications, such as the development of “UroBot,” an AI chatbot capable of handling complex urology-related questions
3. Industry Impact:
   • Organizations are being advised to develop strategic approaches to AI adoption while keeping compliance in mind
   • There’s increasing focus on sector-specific regulations that will affect how different industries can implement AI
   • Companies are working to balance innovation with regulatory requirements
4. Global Perspective:
   • There’s a growing focus on international cooperation and competition in AI development
   • Different regions are taking varied approaches to regulation, with some focusing on consumer protection while others prioritize innovation

Would you like me to explore any of these areas in more detail? I can search for more specific information about any particular aspect that interests you.

Another Sample Session using Brave Search and Claude Desktop**RH:**Can you find pizza restaurants near me?Claude:

*I notice you’d like to find pizza restaurants, but I can’t determine your location from our conversation. To find pizza places near you, I’ll need you to specify a location - it could be a city, neighborhood, landmark, or address. Could you please provide that information?*RH:I live in Manor TexasClaude:

I’ll search for pizza restaurants in Manor, Texas using the local search function.

View Result frombrave_web_searchfrombrave-search(local)

Based on the search results, here are some pizza options in Manor, TX:

1. GRAZIE Pizzeria - A local pizzeria that’s highly rated on Yelp
2. DoubleDave’s Pizzaworks
   • Location: Manor location
   • Known for: Peproni Rolls®, Philly Cheesesteak Stromboli, and specialty pizzas
   • Services: Dine-in, Delivery, and Carry-Out available
3. Domino’s Pizza
   • Address: 11211 US Highway 290 E, Ste 400
   • Hours: Open until midnight
   • Services: Delivery and takeout available
   • Menu includes pizza, wings, sandwiches, and salads
4. Pizza Hut
   • Address: 11300 Highway 290 East Bldg 3
   • Hours: 11:00 AM - 11:00 PM
   • Services: Carryout and delivery
5. Little Caesars
   • Address: 12836 Hwy 290
   • Phone: (737) 357-6622

Each of these places offers different specialties and service options. Would you like more specific information about any of these restaurants, such as their menus, pricing, or customer reviews? I can also search for other local pizzerias if these don’t meet your preferences.

Sequence Diagram: Brave Search MCP Server Integration

The following sequence diagram illustrates how the Brave Search MCP server processes requests through two sample sessions. It shows the flow of communication between theHost,Client, andServercomponents.

Sequence Diagram: Communication Flow

sequenceDiagram
    participant Host as Host (Claude Desktop)
    participant Client as Client (Brave MCP Integration)
    participant Server as Brave Search MCP Server

    Note over Host: Session 1: Web Search

    Host->>Client: Request: "Search for recent news about AI"
    Client->>Server: brave_web_search(query="recent news about AI", count=5, offset=0)
    Server-->>Client: Response: Search results with articles
    Client-->>Host: Deliver search results to Claude interface
    Host->>Client: Request: "Summarize recent developments"
    Client->>Server: brave_web_search(query="recent AI developments", count=5, offset=0)
    Server-->>Client: Response: Summarized updates
    Client-->>Host: Return summary for user

    Note over Host: Session 2: Local Search

    Host->>Client: Request: "Find pizza restaurants near Manor, TX"
    Client->>Server: brave_local_search(query="pizza restaurants", location="Manor, TX", count=5)
    Server-->>Client: Response: List of local restaurants
    Client-->>Host: Return restaurant list to Claude interface
    Host->>Client: Request: "What are their ratings and hours?"
    Client->>Server: brave_local_search(query="pizza restaurants", location="Manor, TX", count=5, details=true)
    Server-->>Client: Response: Ratings and hours for restaurants
    Client-->>Host: Display detailed results

The sequence diagram above illustrates the dynamic communication flow between system components:

• Users initiate requests through the Claude Desktop interface (Host)
• The MCP Client processes and transforms these requests into standardized API calls
• Brave MCP Server acts as an intermediary, coordinating with external APIs and data sources
• Information flows back through the chain, ultimately presenting results to users

System Architecture: Brave Search MCP Integration

graph TD
    subgraph ClaudeDesktop
        A("Host (Claude Interface)")
        B("Client (MCP Integration)")
    end
    C("Brave Search MCP Server")
    D("Brave Search API")
    E("Local Database/Resources")

    A --> B
    B --> C
    C --> D
    C --> E
```**Block Diagram**:

- Illustrates the modular architecture:
    - The Host (Claude Desktop) provides the user interface and communicates with the Client.
    - The Client manages protocol-specific interactions with the Brave MCP Server.
    - The Brave MCP Server connects with external APIs (Brave Search API) and local resources to retrieve data.

The diagrams above illustrate how the integration brings advanced search capabilities to the Claude Desktop environment.

---


### Finding more information about MCP

You can start building and testing MCP connectors right away. If you're a Claude for Work customer, you can test MCP servers locally and connect Claude to your internal systems and datasets. With Anthropic's recently released developer toolkits, you can immediately begin using the existing servers for Claude Desktop—significantly expanding Claude's capabilities and reach. 

Here's how to get started building with MCP:

- Install pre-built MCP servers through the [Claude Desktop app](https://claude.ai/download)
- Follow Anthropic's [quickstart guide](https://modelcontextprotocol.io/quickstart) to build your first MCP server
- Contribute to the [open-source repositories](https://github.com/modelcontextprotocol) of connectors and implementations

The following MCP Server Plugins are available:

- [**Filesystem**](https://github.com/modelcontextprotocol/servers/blob/main/src/filesystem) - File operations with configurable access controls
- [**GitHub**](https://github.com/modelcontextprotocol/servers/blob/main/src/github) - Repository management, file operations, and GitHub API integration
- [**GitLab**](https://github.com/modelcontextprotocol/servers/blob/main/src/gitlab) - GitLab API, enabling project management
- [**Git**](https://github.com/modelcontextprotocol/servers/blob/main/src/git) - Read, search, and manipulate Git repositories
- [**Google Drive**](https://github.com/modelcontextprotocol/servers/blob/main/src/gdrive) - File access and search capabilities for Google Drive
- [**PostgreSQL**](https://github.com/modelcontextprotocol/servers/blob/main/src/postgres) - Read-only database access with schema inspection
- [**SQLite**](https://github.com/modelcontextprotocol/servers/blob/main/src/sqlite) - Database interaction
- [**Slack**](https://github.com/modelcontextprotocol/servers/blob/main/src/slack) - Channel management and messaging
- [**Memory**](https://github.com/modelcontextprotocol/servers/blob/main/src/memory) - Knowledge graph-based persistent memory system
- [**Puppeteer**](https://github.com/modelcontextprotocol/servers/blob/main/src/puppeteer) - Browser automation and web scraping
- [**Brave Search**](https://github.com/modelcontextprotocol/servers/blob/main/src/brave-search) - Web and local search using Brave's Search API
- [**Google Maps**](https://github.com/modelcontextprotocol/servers/blob/main/src/google-maps) - Location services, directions, and place details
- [**Fetch**](https://github.com/modelcontextprotocol/servers/blob/main/src/fetch) - Web content fetching and conversion for efficient LLM usage

---


## Conclusion

Integrating Brave Search MCP with Claude Desktop expands the horizons of AI-assisted tasks, enabling real-time, contextual insights directly from the web or your local environment. This agentic framework will bring about access to many resources allowing you to create powerful agentic applications. By bridging the gap between AI and diverse data sources, this plugin transforms Claude into a powerful tool for productivity, research, and decision-making.

Embrace the future of AI with MCP, and unlock a smarter way to work, learn, and create agentic applications.

---


## Additional Resources

- [Brave Search API Documentation](https://brave.com/search/api/)
- [Claude Desktop Download](https://claude.ai/download)
- [MCP Specifications and SDKs](https://modelcontextprotocol.io/)

---

[**Rick Hightower**](https://www.notion.so/About-133d6bbdbbea813aa509e9585e6867fe?pvs=21) is a seasoned software architect and technology innovator with over three decades of experience in enterprise software development. A TensorFlow Certified Developer, Rick also completed [a Stanford University specialization in Machine Learning/AI,](https://www.notion.so/About-133d6bbdbbea813aa509e9585e6867fe?pvs=21) showing his dedication to advancing AI and data engineering technologies. Renowned within the Java ecosystem and recognized as a 2017 Java Champion, Rick has authored technical publications, contributed to open-source initiatives, and held leadership positions, including as a former executive at a Fortune 100 technology company.

![50629AAE-64D8-4818-A8A2-92B0321783F1.png](/images/anthropic-s-new-mcp-integration-streamlining-ai-as/50629AAE-64D8-4818-A8A2-92B0321783F1.png)

Recently, Rick has been at the forefront of AI innovation, concentrating on large-scale data engineering and the development of AI-driven solutions. He has engaged in groundbreaking projects such as:

-**AI Document Processing**: Used AWS Bedrock and generative AI to extract data from documents and generate deliverables within minutes, rather than months.
-**RAG-based Expert Systems**: Designed and deployed Retrieval-Augmented Generation (RAG) systems employing LLMs such as ChatGPT and Claude, integrating with Google Cloud for document and code repository indexing.
-**Meeting Tools**: Developed tools for real-time question detection, meeting summarization, and action item extraction, significantly enhancing workplace productivity.
-**AR/VR Scalability**: Collaborated with a startup to scale backend services on Azure Kubernetes Service (AKS), deploying seven services with advanced load balancing and caching mechanisms.
-**Legal AI Automation**: Used artificial intelligence (AI) to analyze legal documents, extracting valuable insights at a reduced cost. This resulted in efficiency gains for clients.

As an active mentor and thought leader, Rick actively participates in conferences and shares his insights on the convergence of AI, data engineering, and enterprise software. His practical approach to solving complex challenges has earned him a strong reputation within the technology community.


## Recent articles by Rick

1. [Articles Overview](https://rick-hightower.notion.site/articles)
2. [Setting up Claude Filesystem MCP](Setting%20up%20Claude%20Filesystem%20MCP%2014dd6bbdbbea806c8acae1f350ad234a.md)
3. [Rick and Chris Review Machine Learning](https://rick-hightower.notion.site/rick-and-chris-review-machine-learning?pvs=25)
4. [Streamlit Adventures Part 5 Article](https://rick-hightower.notion.site/streamlit-adventures-part-5-file-viewer-streamlit-refresh?pvs=25)
5. [Streamlit Part 4: Form Validation Part 2](https://rick-hightower.notion.site/article-streamlit-part-4-form-validation-part-2?pvs=25)
6. [Streamlit Part 3 - Form Validation Part 1](https://rick-hightower.notion.site/article-streamlit-part-3?pvs=25)
7. [Advanced SQL Techniques for ETL](https://rick-hightower.notion.site/advanced-sql-techniques-for-etl?pvs=25)
8. [Streamlit Part 2: Layouts, Components, and Graphs](https://rick-hightower.notion.site/streamlit-part-2-layouts-components-and-graphs?pvs=25)
9. [Conversation About Streamlit While Walking in the Park](https://rick-hightower.notion.site/conversation-about-streamlit-while-walking-in-the-park?pvs=25)
10. [PrivateGPT and LlamaIndex](PrivateGPT%20and%20LlamaIndex%20Revolutionizing%20AI%20Proje%20133d6bbdbbea813e9a77eeff437a145a.md)
11. [OpenAI's Latest Developments](OpenAI%E2%80%99s%20Latest%20Developments%20Reasons%20Why%20You%20Shoul%20133d6bbdbbea81c6b889fcf7ecdd9935.md)
12. [AI-Powered Knowledge Base for Product Managers](AI-Powered%20Knowledge%20Base%20for%20Product%20Managers%20133d6bbdbbea81b6a208f09cad8ae1d2.md)
13. [ChatGPT at Scale on Azure Cloud](ChatGPT%20at%20scale%20Azure%20Cloud%20Provides%20Access%20to%20Ch%20133d6bbdbbea815cb0c6d5a398133068.md)
14. [Prompt Engineering with CoT and Synthetic Prompts Part 2](Prompt%20Engineering%20Getting%20What%20You%20Want%20From%20Chat%20134d6bbdbbea80b3bc45e245060c81af.md)
15. [Understanding LLMs and Using Chain of Thoughts](Understanding%20LLM%20and%20using%20Chain%20of%20Thoughts%20134d6bbdbbea8043be90d741028424e2.md)
16. [Meta's Llama 2 AI Model](Meta's%20Llama%202%20Threatens%20Dominance%20of%20Other%20AI%20Mod%20134d6bbdbbea80f0ad35e3004c998249.md)
17. [ChatGPT Embeddings and HyDE for Improved Search](Using%20ChatGPT,%20Embeddings,%20and%20HyDE%20to%20Improve%20Sea%20134d6bbdbbea8002b17bfc5c7b81a69a.md)
18. Rick wrote all of the [blog posts and articles on the Cloudurable site](https://cloudurable.com/blog/index.html)