Vertex AI Agent Platform

Skill: Vertex AI Model Context Protocol (MCP)

This skill outlines the rules for integrating, configuring, and registering Model Context Protocol (MCP) Servers within the Google Cloud Vertex Agent Platform.

1. Structuring toolspec.json

• The platform requires a strict JSON specification defining the tools exposed by your MCP server.
• Payload Format:

  {
    "tools": [
      {
        "name": "example_tool_name",
        "description": "A highly detailed and unambiguous description of what the tool accomplishes.",
        "inputSchema": {
          "type": "object",
          "properties": {
            "param_1": {
              "type": "string",
              "description": "Clear definition of the input parameter."
            }
          },
          "required": ["param_1"]
        }
      }
    ]
  }
  

• Strictness: All parameters in inputSchema must be strongly typed. Ambiguous types (like empty schema references) will cause the tool registry to fail.

Skill: Vertex AI Agent Platform

This skill covers deployment parameters, packaging structures, and runtime configuration for hosting AI Agents on Google Cloud's Vertex platform.

1. Packaging Standards

• Directory Structure: Keep all Dialogflow CX lifecycle callbacks, JSON specs, and OpenAPI toolset configurations in their designated subdirectories.
• OpenAPI Toolsets: OpenAPI schemas must be packaged into a ZIP bundle alongside the agent definitions when uploading to the Agent Studio.

2. Configuration

• Ensure the .agent_engine_config.json file specifies the correct compute resource scaling parameters.
• Always supply the explicit --region (or --location) parameter when invoking deployment scripts, as defaulting to regional configurations causes Cloud Build failures.