NTS

Agentic AI & Cloud AI Infrastructure

Tool Use, MCP & Agent-to-Agent

The Model Context Protocol is becoming the standard way to connect agents to tools, and agent-to-agent protocols let agents delegate to each other. All three clouds now support both patterns.

Multicloud AI & ERP / Agentic AI & Cloud AI Infrastructure

Tool Use, MCP & Agent-to-Agent: diagram comparing AWS, Azure and Google Cloud
How an agent calls a tool through MCP, and how AWS, Azure and Google Cloud each implement it.

Why it matters

Every agent needs tools, and every custom integration is code to maintain. Open protocols let you expose a tool once and use it from any agent, while a gateway keeps that access authenticated, limited and logged.

How it works

How a governed tool call works

The agent decides it needs a tool and calls it through an MCP gateway. The gateway authenticates the agent and checks policy: is this agent allowed this tool? The MCP server wraps an API, database or application and returns the result. For bigger sub-tasks, an agent can delegate to another agent over A2A.

AWS, Azure and Google Cloud compared

AWSAzureGoogle Cloud
MCP gatewayAgentCore GatewayAPI Management for MCPAgent Gateway
Tool policyAgentCore PolicyAgent 365 and Foundry controlsAgent Registry and Gateway policy
Agent identityAgentCore IdentityEntra agent identitiesAgent Identity
Agent-to-agentA2A support in AgentCoreA2A in Agent FrameworkA2A protocol (open standard)
Turning APIs into toolsGateway from OpenAPI and LambdaAPIM and Logic Apps connectorsApigee API hub

What makes each cloud different

AWS

AgentCore Gateway converts existing APIs and Lambda functions into MCP tools, with tool search so agents find the right one, and AgentCore Identity handles OAuth to third-party services.

Azure

Microsoft has made MCP the default integration layer across Foundry, Microsoft 365 Copilot and Agent 365, with API Management exposing and governing MCP servers.

Google Cloud

Google started the open A2A protocol and supports MCP across its agent stack; Agent Gateway fronts agent traffic and Agent Registry lists approved agents and tools.

Design decisions that matter

  1. 01Expose tools once through MCP

    Reuse across every agent.

  2. 02Gateway in front of every tool

    Authentication, rate limits and logs.

  3. 03Least privilege per agent

    Each agent sees only its tools.

  4. 04Delegate with A2A

    Specialist agents instead of giant ones.

Common mistakes we see

  • Direct tool access without a gateway. No control and no audit.
  • Too many tools per agent. Selection accuracy drops.
  • Unvetted public MCP servers. Supply-chain risk inside your agents.

A practical 30-day plan

  1. Week 1

    List the APIs agents need and their owners.

  2. Week 2

    Wrap the first three as MCP tools.

  3. Week 3

    Put them behind a gateway with policy.

  4. Week 4

    Connect two agents with A2A for a delegated task.

How NTS helps

NTS builds governed tool layers for AI agents: MCP servers for your APIs, gateways with policy and identity, and agent-to-agent architectures on AWS, Azure and Google Cloud.

What you get

  • MCP servers for your key business systems
  • Agent-to-agent connections where useful
  • Authentication and authorization for every tool
  • A catalog of approved tools and servers

Typical outcomes

  • Agents connected safely to real systems
  • Faster delivery of new agent capabilities
  • No unvetted tools in production

How we work with you

  1. 1Assess

    A focused review of your MCP and agent integrations across AWS, Azure and Google Cloud. We look for the gaps we find most often: direct tool access without a gateway, too many tools per agent and unvetted public MCP servers.

  2. 2Design

    A target architecture and roadmap built on the decisions that matter most: expose tools once through MCP, gateway in front of every tool, least privilege per agent and delegate with A2A.

  3. 3Build

    Hands-on delivery alongside your team. We list the APIs agents need and their owners, wrap the first three as MCP tools, put them behind a gateway with policy, and finally connect two agents with A2A for a delegated task.

  4. 4Run and improve

    We maintain tool servers, review new integrations and keep the catalog secure.

Why teams choose NTS for MCP and agent integrations

  • Open standards

    We use MCP and A2A so integrations work across models and platforms.

  • Security for tools

    Every tool call is authorized and logged.

  • Reusable integrations

    Build once, use with every agent.

Book a free 30-minute review

Talk to an NTS architect about your MCP and agent integrations on AWS, Azure and Google Cloud.

Email us 469-469-1965