Amazon Web Services → Microsoft Azure
Azure equivalent of Amazon Bedrock AgentCore
The Azure equivalent of Amazon Bedrock AgentCore is Foundry Agent Service.
About Amazon Bedrock AgentCore
| Provider | Amazon Web Services |
|---|---|
| Status | Generally available |
| What it is | Modular agent platform. Harness (managed agent loop), Runtime, Memory, Gateway (MCP), Identity, Code Interpreter, Browser, Observability, Evaluations, Optimization, Policy, Registry, Payments. Works with any framework and any model. |
| Documentation | docs.aws.amazon.com |
| Pricing | Official pricing page |
Azure mappings (6)
| Capability | Azure equivalent | Match |
|---|---|---|
| Model, RAG & agent evaluation | Foundry observability & evaluations | Direct match |
| Managed AI agent platform | Foundry Agent Service | Direct match |
| Configuration-based agents (legacy Bedrock Agents) | Foundry Agent Service | Partial match |
| Agent tools, MCP gateway & sandboxes | Foundry Agent Service | Partial match |
| Agent memory (short- & long-term) | Foundry Agent Service | Partial match |
| Agent identity, policy & registry | Foundry Agent Service, Foundry guardrails (content filters) | Partial match |
Generative AI platforms & models(1)
Model, RAG & agent evaluation Direct match
Bedrock Evaluations and AgentCore Evaluations map to Foundry observability and evaluations (built-in and custom evaluators, continuous evaluation, and tracing).
- Bedrock Evaluations covers models (including Marketplace, custom, and imported models, and prompt routers) and RAG sources, including ones outside Bedrock. It supports human-worker evaluation jobs.
- AgentCore Evaluations and Optimization score agent sessions and traces (Strands, LangGraph, OpenTelemetry/OpenInference) and run A/B tests on prompt and tool-description changes.
- Foundry combines tracing, evaluators, and monitoring dashboards (some preview), and adds Agent optimizer (preview) to propose better instructions and tools.
AI agents(5)
Managed AI agent platform Direct match
The Azure equivalent of Amazon Bedrock AgentCore is Foundry Agent Service. Both host declarative or code-based agents from any framework and model.
- Shape. AgentCore is a set of modular services you can use independently: Harness (config-defined managed agent loop in an isolated microVM), Runtime (serverless hosting for code agents), Memory, Gateway, Identity, Code Interpreter, Browser, Observability, Evaluations, Optimization, Policy, Registry, and Payments.
- Foundry Agent Service has three agent types. Prompt agents are declarative. Voice-based prompt agents run on Voice Live (preview). Hosted agents run your container or zip. You can also call the Responses API for "ephemeral" agents you define in code.
- Frameworks. AgentCore Runtime lists Strands, LangGraph, CrewAI, LlamaIndex, Google ADK, and OpenAI Agents SDK. Foundry hosted agents list Microsoft Agent Framework, LangGraph, OpenAI Agents SDK, Anthropic Agent SDK, and GitHub Copilot SDK.
Configuration-based agents (legacy Bedrock Agents) Partial match
Bedrock Agents Classic is in maintenance (no new customers since July 30, 2026). On Azure, the equivalent concept is a Foundry prompt agent, and on AWS the successor is the AgentCore managed harness.
- AWS's migration path is the AgentCore harness. Action groups become MCP tools through AgentCore Gateway, and knowledge bases become gateway-fronted tools, with memory, guardrails, and tracing added.
- Foundry prompt agents are the declarative counterpart. They support instructions, a model, and tools (file search, code interpreter, web search, MCP, functions).
Agent tools, MCP gateway & sandboxes Partial match
AgentCore Gateway, Code Interpreter, and Browser map to the Foundry Agent Service toolbox (a managed MCP endpoint with web search, file search, code interpreter, OpenAPI, functions, and MCP tools).
- AgentCore Gateway turns REST APIs, Lambda functions, and existing services into MCP tools, and fronts existing MCP servers with interceptors and Cedar-compatible policy.
- AgentCore Browser is a managed cloud browser runtime for agents (Playwright and BrowserUse compatible). It's a standalone service, not just a tool flag.
- The Foundry toolbox curates tools once and shares them across agents through a single managed MCP endpoint with central auth and versioning. It also offers Microsoft-ecosystem tools (SharePoint, WorkIQ, Fabric IQ).
Agent memory (short- & long-term) Partial match
AgentCore Memory maps to Foundry Agent Service memory (preview). AgentCore's memory is GA and more configurable today.
- AgentCore Memory offers short-term (multi-turn) and long-term memory with configurable strategies, sharing across agents, and namespace-scoped IAM controls.
- Foundry lists memory as a built-in tool and describes it as preview on the Foundry overview. Hosted agents also get session-level state persistence.
Agent identity, policy & registry Partial match
AgentCore Identity, Policy, and Registry map to Microsoft Entra agent identities, Foundry guardrails (tool-call controls), the Foundry control plane, and the Entra Agent Registry.
- AgentCore Identity vends short-lived credentials from a token vault, and works with any IdP (Cognito, Okta, Entra ID, Auth0).
- AgentCore Policy enforces deterministic rules on every tool call at the Gateway, written in natural language or Dogwood (Cedar-compatible).
- On Azure, hosted agents get a dedicated Microsoft Entra identity. Guardrails can inspect tool calls and tool responses (preview), and hosted agents support network egress controls (preview).