Explore agent integrations and multiagent solutions
Grounding gives an agent knowledge. Actions give it agency. The previous unit covered how agents get reliable access to organizational data. This one covers the other side: how they reach out to enterprise systems, external APIs, and other agents.
Connect agents to enterprise systems
Agents often need to take action in enterprise systems: submitting a request, creating a record, initiating a workflow, or triggering a notification. Microsoft's platform provides integration options across all development platforms, including the open standard Model Context Protocol (MCP), which works across all of them.
MCP is an open standard for connecting AI models to external tools and data sources. An agent can call tools exposed through an MCP server, and any system that exposes an MCP endpoint becomes callable by MCP-compatible agents. On Microsoft's platform, MCP is one of the primary ways agents connect to external tools. The IQ services from the previous unit all support MCP as a query interface.
Connect agents in Copilot Studio
In Copilot Studio, agents access external systems through a library of prebuilt Power Platform connectors, Power Automate flows, and direct REST API calls.
- Power Platform connectors provide managed, authenticated access to a wide range of enterprise applications without requiring custom API code. Custom connectors extend this to proprietary or third-party systems.
- Power Automate flows let the agent trigger structured workflows: Multistep process sequences that run independently of the conversation and call back when complete.
- Direct REST API actions provide explicit HTTP integration for systems that don't have a connector.
- Computer Use enables agents to interact with systems that have no API (legacy applications, web-based tools, and desktop software) by simulating user input through vision-based interaction.
Connect agents using code
For pro-code agents, integrations are implemented as registered tools that the orchestration framework calls when the agent's reasoning determines they're needed.
Azure Logic Apps provides 1,400+ managed connectors and can be called as an agent tool through its built-in HTTP trigger, acting as integration middleware for authentication, retry logic, and connection management.
Azure Functions offers lightweight, event-driven compute for calling custom APIs or handling integrations that don't fit a managed connector.
OpenAPI-defined APIs can be registered as tools directly, making any REST API callable through the same tool-dispatch mechanism as framework-provided tools.
Computer Use is also available for pro-code agents (in preview at time of authoring), using the same vision-based approach to interact with systems that have no API.
Build with multiple agents
For some scenarios, a single agent isn't the right architecture. Multiagent solutions distribute specialized work across a coordinated network, with each agent handling the tasks it's optimized for.
Within Microsoft's platform, multiagent connections use platform-native paths: Copilot Studio connects to Foundry agents through a direct native connection, and Foundry agents coordinate internally through Workflows. The Agent-to-Agent (A2A) protocol, an open standard for agent communication, is a mechanism for cross-agent coordination when agents are built on different frameworks and platforms. Consider using A2A to integrate your agents with external agents hosted outside Microsoft's platform or owned by different organizations.
A Foundry Hosted agent can also expose an incoming A2A endpoint (in preview at time of authoring), making it callable by any A2A-compatible agent regardless of framework.
Trust is a first-class concern in multiagent scenarios. Cross-tenant agent coordination requires explicit trust configuration; it isn't established automatically.
Guiding question: What external systems will the agent in your scenario need to act on? And does that scenario involve coordinating with other agents?