Skip to main content

MCP 概览

快速链接​

概览​

模型上下文协议是一个开放协议，标准化了应用程序为 LLM 提供上下文的方式。将 MCP 想象成 AI 应用程序的 USB-C 端口；它提供了将 AI 模型连接到不同数据源和工具的标准化方式。MCP 服务器充当大语言模型（LLM）（如 Claude）与外部工具或数据源之间的中介。它们是向 LLM 公开功能的小程序，使它们能够通过 MCP 与外部世界交互。MCP 服务器本质上就像 LLM 可以使用的 API。

MCP diagram showing how MCP servers connect LLMs to external tools and data sources

Key Concepts​

MCP servers define a set of "tools," which are functions the LLM can execute. These tools offer a wide range of capabilities.

Here's how MCP works:

  • MCP hosts discover the capabilities of connected servers and load their tools, prompts, and resources.
  • Resources provide consistent access to read-only data, akin to file paths or database queries.
  • Security is ensured as servers isolate credentials and sensitive data. Interactions require explicit user approval.

Use Cases​

The potential of MCP servers is vast. They can be used for a variety of purposes.

Here are some concrete examples of how MCP servers can be used:

  • Web Services and API Integration:
    • Monitor GitHub repositories for new issues
    • Post updates to Twitter based on specific triggers
    • Retrieve real-time weather data for location-based services
  • Browser Automation:
    • Automate web application testing
    • Scrape e-commerce sites for price comparisons
    • Generate screenshots for website monitoring
  • Database Queries:
    • Generate weekly sales reports
    • Analyze customer behavior patterns
    • Create real-time dashboards for business metrics
  • Project and Task Management:
    • Automate Jira ticket creation based on code commits
    • Generate weekly progress reports
    • Create task dependencies based on project requirements
  • Codebase Documentation:
    • Generate API documentation from code comments
    • Create architecture diagrams from code structure
    • Maintain up-to-date README files

Getting Started​

Caret does not come with any pre-installed MCP servers. You'll need to find and install them separately.

Choose the right approach for your needs:

  • Community Repositories: Check for community-maintained lists of MCP servers on GitHub. See Adding MCP Servers from Github
  • Caret Marketplace: Install one from Caret's MCP Marketplace
  • Ask Caret: You can ask Caret to help you find or create MCP servers
  • Build Your Own: Create custom MCP servers using the MCP SDK
  • Customize Existing Servers: Modify existing servers to fit your specific requirements

Integration with Caret​

Caret simplifies the building and use of MCP servers through its AI capabilities.

Building MCP Servers​

  • Natural language understanding: Instruct Caret in natural language to build an MCP server by describing its functionalities. Caret will interpret your instructions and generate the necessary code.
  • Cloning and building servers: Caret can clone existing MCP server repositories from GitHub and build them automatically.
  • Configuration and dependency management: Caret handles configuration files, environment variables, and dependencies.
  • Troubleshooting and debugging: Caret helps identify and resolve errors during development.

Using MCP Servers​

  • Tool execution: Caret seamlessly integrates with MCP servers, allowing you to execute their defined tools.
  • Context-aware interactions: Caret can intelligently suggest using relevant tools based on conversation context.
  • Dynamic integrations: Combine multiple MCP server capabilities for complex tasks. For example, Caret could use a GitHub server to get data and a Notion server to create a formatted report.

Security Considerations​

When working with MCP servers, it's important to follow security best practices:

  • Authentication: Always use secure authentication methods for API access
  • Environment Variables: Store sensitive information in environment variables
  • Access Control: Limit server access to authorized users only
  • Data Validation: Validate all inputs to prevent injection attacks
  • Logging: Implement secure logging practices without exposing sensitive data

Resources​

There are various resources available for finding and learning about MCP servers.

Here are some links to resources for finding and learning about MCP servers: