Dynamic MCP tool schema issue

[Sagnnik]

I was testing the dynamic mcp tools with gpt-5-mini. When I tried to look deeper, I noticed that the tool schemas are not what I expected.

Example:

[
   {
      "description":"Create a JavaScript-enabled tool that combines multiple MCP server tools. \nThis allows you to write scripts that call multiple tools and combine their results.\nUse the mcp-find tool to find servers and make sure they are are ready with the mcp-add tool. When running\nmcp-add, we don't have to activate the tools.\n",
      "inputSchema":{
         "type":"object",
         "required":[
            "servers",
            "name"
         ],
         "properties":{
            "name":{
               "type":"string",
               "description":"Name for the new code-mode tool (will be prefixed with 'code-mode-')"
            },
            "servers":{
               "type":"array",
               "description":"List of MCP server names whose tools should be available in the JavaScript environment",
               "items":{
                  "type":"string"
               }
            }
         }
      },
      "name":"code-mode"
   },
   {
      "description":"Execute a tool that exists in the current session. This allows calling tools that may not be visible in listTools results.",
      "inputSchema":{
         "type":"object",
         "required":[
            "name"
         ],
         "properties":{
            "arguments":{
               "type":[
                  "string",
                  "number",
                  "boolean",
                  "object",
                  "array",
                  "null"
               ],
               "description":"Arguments to pass to the tool (can be any valid JSON value)"
            },
            "name":{
               "type":"string",
               "description":"Name of the tool to execute"
            }
         }
      },
      "name":"mcp-exec"
   },
   {
      "description":"Find MCP servers in the current catalog by name, title, or description. Returns matching servers with their details.",
      "inputSchema":{
         "type":"object",
         "required":[
            "query"
         ],
         "properties":{
            "limit":{
               "type":"integer",
               "description":"Maximum number of results to return (default: 10)"
            },
            "query":{
               "type":"string",
               "description":"Search query to find servers by name, title, or description (case-insensitive)"
            }
         }
      },
      "name":"mcp-find"
   },
]

I am exposing these 3 tools to the LLM. But why are the arguments which are needed not mentioned for the LLM? Am I missing something?
Like I looked into the source code and code-mode needs code, servers, name, timeout and mcp-exec needs name from code-mode-{name} and script args. It does work with good system prompt however, but that's not reliable right?