DEV Community

Cover image for How to Connect an AI Tool to a Local MCP Server: A Practical Example
Emerson Skaggs
Emerson Skaggs

Posted on Originally published at bitbrowser.net

How to Connect an AI Tool to a Local MCP Server: A Practical Example

#ai #automation #browser #mcp

Model Context Protocol (MCP) is making it easier for AI assistants to interact with external tools and applications.

Instead of manually clicking through every setting, you can connect an MCP server to a compatible AI tool, describe what you want in natural language, and let the assistant handle the individual operations.

A practical example is BitBrowser 7.1.5, which added MCP support through its local API.

This article walks through the setup and shows how MCP can be used for everyday browser management tasks.

What You Need

For this example, you need:

  • BitBrowser 7.1.5 or later
  • An AI tool that supports MCP
  • BitBrowser's Local API enabled
  • An MCP configuration generated by BitBrowser

The same basic approach can also be applied to other applications that expose their functionality through MCP.

1. Enable the Local MCP Server

Open BitBrowser and go to:

Settings → Browser Settings → Local API

Enable Authentication Control.

Enable Authentication Control

BitBrowser will generate an API token and display the required request header:

x-api-key: <YOUR_API_TOKEN>
Enter fullscreen mode Exit fullscreen mode

The default endpoints are:

Local API: http://127.0.0.1:54345
MCP: http://127.0.0.1:54345/mcp
Enter fullscreen mode Exit fullscreen mode

If you use a different Local API port, use the address shown in your settings.

2. Get the MCP Configuration

Scroll down to the MCP section.

BitBrowser provides configurations for several compatible AI tools, including:

  • Codex
  • Cursor
  • OpenCode
  • WorkBuddy
  • Qwen
  • Hermes
  • OpenClaw
  • General
  • Claude Code

Select your AI tool and copy the generated configuration.

Select your AI tool

Using the generated configuration is usually the simplest option because the endpoint and authentication settings are already included.

generated configuration

For a generic MCP client, the configuration looks similar to:

{
  "mcpServers": {
    "bitbrowser": {
      "url": "http://127.0.0.1:54345/mcp",
      "headers": {
        "x-api-key": "<YOUR_API_TOKEN>"
      }
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Replace

<YOUR_API_TOKEN>
Enter fullscreen mode Exit fullscreen mode

with your actual token.

Keep the x-api-key header name unchanged.

3. Connect the MCP Server

Open your AI tool's MCP settings.

Depending on the application, this may be called:

  • MCP Servers
  • Custom MCP
  • Tools
  • Integrations
  • Connectors

Add a new MCP server and paste the configuration.

After saving, check that the server is enabled and that its tools are available.

Add a new MCP server

check that the server is enabled

4. Start with a Read-Only Request

Before asking the AI to make changes, test the connection with a read-only request.

For example:

Show me my current BitBrowser setup.

List all existing Browser Profiles and cloud phones, including their names, groups, and status.

Do not make any changes.
Enter fullscreen mode Exit fullscreen mode

If the AI can retrieve the information correctly, the MCP connection is working.

Starting with a read-only request is also a good way to avoid accidental changes while testing a new integration.

5. Create Browser Profiles with Natural Language

Once the connection works, you can ask the AI to perform actual operations.

For example:

Create 3 new Browser Profiles named MCP_Demo_01, MCP_Demo_02, and MCP_Demo_03.

Use English (US) as the browser language.

Create a group called "MCP Demo" and add all three profiles.

Do not modify any existing profiles.
Enter fullscreen mode Exit fullscreen mode

Verify that all three profiles were created successfully and added to the correct group.

Verify that all three profiles were created successfully

A single request can cover several operations:

Create profiles → configure settings → create a group → assign profiles → verify

You don't have to manually select each underlying MCP tool.

6. Configure Proxies

MCP can also be used for proxy configuration.

For example:

Configure the following SOCKS5 proxies for the three profiles in the "MCP Demo" group.

MCP_Demo_01 → proxy 1
MCP_Demo_02 → proxy 2
MCP_Demo_03 → proxy 3

Keep all other Browser Profile settings unchanged.

Verify each proxy after updating.
Enter fullscreen mode Exit fullscreen mode

A useful pattern is to clearly specify two things:

What should change

and

What should remain unchanged

This reduces the chance of unintended modifications.

Never include real proxy passwords, API tokens, or other sensitive credentials in public prompts or examples.

7. Make Targeted Changes

MCP doesn't have to be used for large workflows.

It can also handle small maintenance tasks.

For example:

Change the browser language of MCP_Demo_01 and MCP_Demo_02 to French.

Keep their proxy settings, group assignments, and other fingerprint settings unchanged.

Verify the changes after updating.
Enter fullscreen mode Exit fullscreen mode

Another example:

Randomize the browser fingerprints for MCP_Demo_02 and MCP_Demo_03.

Do not change their proxy settings, language, or group assignments.
Enter fullscreen mode Exit fullscreen mode

This makes natural language useful for both routine maintenance and more complex workflows.

8. Combine Multiple Operations

The more interesting use case is combining several operations into one request.

For example:

Create 5 new Browser Profiles named Project_01 through Project_05.

Use English (US) as the browser language.

Create a group called "Project Profiles" and add all five profiles.

Configure the proxy assigned to each profile.

Verify that all five profiles exist, are in the correct group, have the correct proxy, and use English (US).

Do not modify existing profiles outside this group.
Enter fullscreen mode Exit fullscreen mode

The AI can break this request into multiple operations and then check the resulting state.

The basic idea is simple:

Describe the desired result. Let the AI handle the individual operations.

Troubleshooting

If the AI cannot connect, check the following:

  • BitBrowser is running
  • BitBrowser is version 7.1.5 or later
  • Local API is enabled
  • Authentication Control is enabled
  • The MCP endpoint is correct
  • x-api-key is included
  • The API token is correct
  • The MCP server is enabled in your AI tool

If the AI connects but an operation fails, start by asking it to inspect the relevant item without making changes.

For example:

Check the current status of MCP_Demo_01.
Do not make any changes.

If the AI reports that an operation succeeded but the interface does not appear to have changed, refresh the application and query the item again before repeating the operation.

Security Considerations

Local MCP integrations still involve credentials and access permissions.

Keep the following private:

  • API tokens
  • Proxy usernames and passwords
  • Other authentication credentials
  • Private configuration files

Don't publish them in screenshots, repositories, public prompts, or social posts.

It's also a good practice to begin with read-only requests and explicitly tell the AI what it should not modify.

Final Thoughts

MCP adds a different way to interact with software.

Instead of navigating through every routine operation manually, you can describe the task and let an AI assistant coordinate the available tools.

The BitBrowser MCP integration is one practical example of this approach, covering browser profiles, proxies, fingerprints, groups, cloud phones, and other management operations.

If you're trying MCP for the first time, keep it simple:

Connect → test with a read-only request → try one small task → build larger workflows.

Top comments (0)