Skip to main content

Using MCP Tools

MCP tools from configured servers are automatically available to agents during execution.

Tool Discovery​

Tools are automatically discovered when MCP servers are initialized:

# List available tools
rad mcp tools

# List tools from specific server
rad mcp tools --server database-server

Tool Naming​

Tools use the format server-name:tool-name to avoid conflicts:

  • If a tool name is unique, it's available as tool-name
  • If there's a conflict, tools are prefixed: server-name:tool-name

Example:

database-server:query
api-server:query

Both servers have a query tool, so they're prefixed with server names.

Using Tools in Agents​

MCP tools are automatically available to agents. No special configuration is needed.

Example agent execution:

{
  "id": "data-agent",
  "name": "Data Agent",
  "description": "Agent that uses database tools",
  "prompt": "prompts/data-agent.md"
}

When this agent executes, it can use MCP tools like database-server:query just like built-in tools.

Tool Execution​

Tools are executed through the MCP protocol:

  1. Agent requests tool execution
  2. Radium routes to appropriate MCP server
  3. Tool executes on remote server
  4. Results returned to agent

Rich Content Support​

MCP tools can return rich content:

  • Text: Displayed inline
  • Images: Saved to temp files, paths shown
  • Audio: Saved to temp files, paths shown

Example output:

Tool executed successfully
[Image: image/png] Saved to: /tmp/radium_mcp_abc123.png

Error Handling​

MCP tool errors are handled gracefully:

  • Connection errors are logged
  • Tool execution errors are returned to agent
  • Failed servers don't block other servers

Tool Schemas​

Tool input schemas are automatically converted from MCP JSON Schema format to agent-compatible format. Agents receive proper schema information for validation.

Examples​

See agent-with-mcp-tools.toml for an example agent configuration.