Cloudera Docs

Model Context Protocol (MCP) integration guide

MCP enhances AI workflows by enabling seamless integration with external systems and tools. It provides a standardized approach for secure communication, allowing AI agents to interact efficiently with diverse services while maintaining flexibility and control within Cloudera AI Agent Studio.

MCP is an open standard that allows AI applications to provide Large Language Models (LLMs) with context drawn from external data sources and tools. By enabling AI agents to establish secure connections with these external systems, MCP significantly expands the functional scope of your AI workflows.

In Agent Studio, MCP servers act as tool families for specific services, such as Slack, Jira, databases, and web search, facilitating secure and controlled agent interaction with these external environments.

Key features of MCP servers

  • MCP servers as tool families - MCP servers group related tools that grant access to specific services or data sources, for example:
    • DuckDuckGo MCP server: For web search and content retrieval.
    • Slack MCP server: For managing channels, reading messages, and posting updates.
    • Jira MCP server: For querying issues, creating tickets, and updating status.
    • Database MCP server: For inserting, updating, and querying database records.
  • Tool discovery - Agent Studio automatically detects available tools by referencing registered MCP servers. In cases when some tools are not automatically detected —an issue that might occasionally occur —the MCP server remains fully functional and can still be used within workflows
  • Supported MCP server types - Agent Studio supports the execution of MCP servers using environment-specific tools for better isolation and dependency management, eliminating the need for global installation. The supported execution methods are:
    • Python-based servers: Executed using the uvx command.
    • Node.js-based servers: Executed using the npx command.
  • Transport limitations - Agent Studio currently supports only local MCP servers that utilize the standard Input/Output (stdio) transport mechanism, where supported MCP servers operate as local processes, communicating through stdin or stdout streams.

For a detailed understanding of MCP transport mechanisms, refer to the MCP Transport Documentation.