Skip to main content

Configure MCP Services

info

Before reading this chapter, please review the official MCP protocol documentation and obtain the services you need.

For development, you can refer to:

Official docs
Time MCP Server

We recommend following the approach used in the official tutorials when building MCP services. If you use Python, prefer the official package.

Configure Local MCP Services​

warning

Local MCP services have two characteristics:

  1. They are installed on the local computer, and the same MCP configuration is shared across different users who sign in on that machine.
  2. Step editing is not supported. If you need a shareable setup, prefer MCP services configured in the console.

1. Obtain an MCP Server​

Before configuring bit-Agent, download and deploy an MCP server first.

You can refer to the official MCP examples and choose a suitable service, for example:

  • filesystem: access to the local file system
  • sqlite: access an SQLite database through MCP
  • custom: your own custom capability service

2. Configure the MCP Server in bit-Agent​

  1. Open the bit-Agent application.
  2. Click the Settings icon in the upper-right corner.
  3. In Settings, open Service Configuration.
  4. In the MCP Server section, enter the configuration according to the MCP service protocol. The example below configures the official filesystem and sqlite services. Fill in the paths according to your actual environment:
{
  "mcpServers": {
    "filesystem": {
      "command": "cmd",
      "args": [
        "/c",
        "npx",
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "D:\\agent\\data"
      ]
    },
    "sqlite": {
      "command": "cmd",
      "args": [
        "/c",
        "uvx",
        "--directory",
        "D:\\agent\\servers-main\\src\\sqlite",
        "mcp-server-sqlite",
        "--db-path",
        "D:\\agent\\data\\test.db"
      ]
    }
  }
}
  1. After configuration, click Save. bit-Agent will automatically start the corresponding services. A successfully started service shows the status Normal; otherwise it shows Abnormal. Services in an abnormal state cannot be used.

  1. Click Tools to view the tools exposed by that service.

  1. In Settings, you can choose which modes the service is available in. By default, it can be used in both exploration mode and general mode.

Auto-Loading from bit-Crew​

Starting from V3.3, if you have already configured and enabled MCP services in bit-Crew, bit-Agent can load those available services automatically, without requiring duplicated configuration in the client.

Typical Use Cases​

  • Your team manages MCP services centrally in bit-Crew
  • You want to reuse the same service configuration across multiple endpoints

Recommendations​

  • On first use, confirm in a conversation that the tools have appeared
  • If they do not appear automatically, check account permissions and service status, then try again

What is MCP?​

MCP (Model Control Protocol) is an open-protocol standard for providing large language models with secure, extensible access to external tools and data.

Through MCP, AI applications such as intelligent agents, IDE plugins, or enterprise AI platforms can communicate with multiple external services in a standard way, enabling access to:

  • Local file systems, such as reading and writing files
  • Databases, such as SQLite or PostgreSQL
  • Remote API services, such as third-party business systems
  • Internal enterprise services, such as knowledge bases or permission systems

With MCP, AI models can do more than understand and reason. They can safely execute real operations, such as querying databases, modifying configuration, or analyzing log files, which significantly expands both perception and execution boundaries.

How MCP Works​

The MCP protocol defines the communication standard between models and external services.

It usually involves three main roles:

  1. MCP Server: a service that provides a certain capability or resource, such as a file system, database, or Web API
  2. MCP Client: the AI application that initiates requests, such as bit-Agent, an IDE plugin, or an LLM inference framework
  3. MCP Stream: the communication channel between the model and the service, typically based on WebSocket or Streamable HTTP

The interaction flow is:

  • The client sends a request, such as "read file contents" or "run an SQL query"
  • The server receives the request and executes the task
  • The server returns a structured response
  • The model continues making decisions or generating output based on that response