MCP Tool Integration
Explore how the Model Context Protocol standardizes connections between agents and enterprise tools and data. Learn when to use managed or custom MCP servers, how to design narrow tools, and why permissions must come from the authenticated session rather than the model.
An enterprise agent is useful only when it can reach the systems where the customer’s work happens: databases, ticketing systems, document stores, and internal APIs. Before a standard existed, each connection between an agent and a system required custom integration code, and that code was rarely reusable across frameworks or projects. For an FDE moving between customers, this meant rebuilding similar connectors repeatedly and reviewing each one separately for security.
The Model Context Protocol (MCP) addresses this problem by defining a common way for AI applications to discover and call external tools. Google Cloud supports MCP across its agent stack, offers managed MCP servers for its own services, and lets ADK agents connect to any MCP server. FDE interviewers expect candidates to explain how MCP works, when to use a managed server versus a custom one, and, most importantly, how permissions are enforced when an agent acts through these tools.
What MCP is
MCP is an open protocol, introduced by Anthropic in late 2024 and now widely adopted, that standardizes how an AI application connects to external capabilities. It follows a client-server model. The MCP client lives inside the AI application or agent framework, such as ADK. The MCP server wraps a system, such as a database or an API, and exposes its capabilities in a standard format.
An MCP server can expose three kinds of capabilities.
Capability | What it provides | Example |
Tools | Actions the agent can invoke | Query a table, create a ticket, send a notification |
Resources | Read-only data the application can load as context | A file, a database schema, a policy document |
Prompts | Reusable prompt templates offered by the server | A standard template for summarizing an incident |
Tools are the capability FDEs work with most often. When an agent connects to a server, it asks which tools are ...