--- title: "Microsoft Agent Framework vs. Other Agent Frameworks" url: "https://atlan.com/know/ai-agent/microsoft/microsoft-agent-framework-vs-other-agent-frameworks/" description: "Compare Microsoft Agent Framework with Strands, Google ADK, OpenAI Agents SDK, Claude Agent SDK, LangGraph, and CrewAI on languages, protocols, and hosting." author: "Kovid Rathee" author_role: "Head of Solution Architecture, Data & AI, Nexifi" published: "2026-10-06" updated: "2026-10-06" --- > Atlan is hosting Context Conference, bringing together the leaders and builders at the frontier of giving AI the context it needs to understand their business. It runs online on October 28, 2026, from 11:00 AM to 2:00 PM ET. Atlan co-founder Prukalpa Sankar opens and closes the day. Leaders from AstraZeneca, BNY and Verizon share why they invest in context and what they get from it. Registrants get early access to The AI Context Gap, a new study from MIT Technology Review Insights. Register: https://atlan.com/context-conference/ --- Every framework in this comparison runs agents, and each one still needs to know what your enterprise data means. Atlan's [enterprise context layer](https://atlan.com/know/what-is-the-enterprise-context-layer/) supplies that through the Enterprise Data Graph, lineage, and policies, and agents reach it through the [**Atlan MCP Server**](https://atlan.com/know/what-is-atlan-mcp/). Pick the framework for your language, cloud, and models, then connect each enterprise system to the context layer once. ### Pick an Agent Framework for Your Team Takes what you are building, your control flow, language, deployment model, governance needs, and state requirements, and returns a ranked shortlist of three frameworks with the trade-off behind each rank and what breaks first. [Read the skill](/skills/agent-harness-picker.md). *Paste into a new chat* ``` Use the skill at https://atlan.com/skills/agent-harness-picker.md to pick the right agent framework for our team. Ask me for whatever it needs. ``` *Run once in a terminal* ``` curl -fsSL --create-dirs \ -o ~/.agents/skills/agent-harness-picker/SKILL.md \ https://atlan.com/skills/agent-harness-picker.md ``` *For an agent* ``` curl -fsSL https://atlan.com/skills/agent-harness-picker.md ``` Human Agent ClaudeChatGPTGeminiCursorOther Copy .skp{margin:1.75rem 0;padding:1rem 1.125rem 1.125rem} .skp h3{margin:0 0 .375rem!important;font-size:20px!important;line-height:24px!important; font-family:var(--font-funnel-display)!important;font-weight:600!important;color:#2B2B39!important} .skp > p{margin:0 0 .875rem!important;font-size:14px!important;line-height:22px!important;color:#555572!important} .skp > p a{color:#2026D2;text-decoration:underline} .skp-box{margin-bottom:.75rem} .skp-p em{display:block;font-size:10px;font-weight:700;letter-spacing:.06em;text-transform:uppercase; color:#F34D77;font-style:normal;margin-bottom:.25rem} .skp-p > p{margin:0!important} #seo-layout article .skp .skp-p pre{margin:0!important;overflow-x:auto} #seo-layout article .skp .skp-p pre code{color:#2B2B39!important;font-size:12px!important;line-height:19px; white-space:pre-wrap;overflow-wrap:anywhere;background:none;padding:0; font-family:ui-monospace,SFMono-Regular,Menlo,monospace} .skp-p + .skp-p{margin-top:.625rem} .skp-cs{display:flex;flex-wrap:wrap;align-items:center;gap:.375rem} .skp-seg{display:inline-flex;flex:0 0 auto;border:1px solid #DDDDE3;border-radius:999px;overflow:hidden;background:#fff} .skp-m{min-height:32px;padding:.25rem .75rem;border:0;background:transparent;font-family:inherit;color:#77778E;cursor:pointer} .skp-m:hover{color:#2026D2} .skp-m.on{background:#2026D2;color:#fff} .skp-sep{flex:0 0 auto;width:1px;height:20px;background:#DDDDE3;margin:0 .25rem} .skp-tools{display:inline-flex;flex-wrap:wrap;gap:.375rem;min-width:0} .skp-cp{margin-left:auto;min-height:32px;padding:.25rem .75rem;border-radius:999px;background:#fff; font-family:inherit;color:#555572;cursor:pointer} .skp-cp:hover{border-color:#2026D2!important;color:#2026D2} .skp-cs .skp-c{display:inline-flex!important;width:auto;flex:0 0 auto;align-items:center;gap:.375rem; min-height:32px;padding:.25rem .625rem;border-radius:999px;background:#fff;font-family:inherit; color:#555572;cursor:pointer} .skp-cs .skp-c img{display:block!important;width:14px!important;height:14px!important; max-width:14px!important;flex:0 0 auto;opacity:.55;margin:0} .skp-c:hover img,.skp-c.on img{opacity:1} .skp-cs .skp-c:hover{border-color:#797DE4!important;color:#2026D2} .skp-cs .skp-c.on{border-color:#2026D2!important;background:#F4F4FD;color:#2026D2} .skp-agent .skp-tools,.skp-agent .skp-sep{display:none} (function(){ var p=document.querySelector('.skp[data-skill="agent-harness-picker"]');if(!p)return; var panes=p.querySelectorAll('.skp-p'); function show(id){panes.forEach(function(x){x.hidden=x.dataset.skp!==id;});} function cur(){var a=p.querySelector('.skp-p:not([hidden]) pre code');return a?a.textContent:'';} show('prompt'); p.querySelectorAll('.skp-p em').forEach(function(e){ var par=e.parentElement; ((par.tagName==='P'&&par.textContent.trim()===e.textContent.trim())?par:e).hidden=true; }); p.querySelectorAll('.skp-c').forEach(function(b){ b.addEventListener('click',function(){ p.querySelectorAll('.skp-c').forEach(function(x){var on=x===b;x.classList.toggle('on',on);x.setAttribute('aria-pressed',on);}); show(b.dataset.skp); }); }); p.querySelectorAll('.skp-m').forEach(function(m){ m.addEventListener('click',function(){ var agent=m.dataset.skpMode==='agent'; p.classList.toggle('skp-agent',agent); p.querySelectorAll('.skp-m').forEach(function(x){var on=x===m;x.classList.toggle('on',on);x.setAttribute('aria-pressed',on);}); if(agent){show('agent');}else{var s=p.querySelector('.skp-c.on');show(s?s.dataset.skp:'prompt');} }); }); var c=p.querySelector('[data-skp-copy]'); c.addEventListener('click',function(){ navigator.clipboard.writeText(cur()).then(function(){ c.textContent='Copied';setTimeout(function(){c.textContent='Copy';},1600); }); }); })(); The frameworks differ across four dimensions: - **Cloud pairing:** Microsoft Agent Framework, AWS Strands, and Google ADK each pair with one cloud's managed hosting, and all three accept models from other providers. - **Lab runtimes:** OpenAI Agents SDK and Claude Agent SDK come with the runtime scaffolding of the labs that build them. - **Independent stacks:** LangGraph and CrewAI run on any cloud and bring their own managed platforms. - **Language and protocols:** only Microsoft Agent Framework lists .NET, and all seven frameworks document MCP support. | | | |---|---| | **What it is** | A comparison of Microsoft Agent Framework with six other agent frameworks, grouped by who builds them | | **Key trade-off** | Cloud-paired hosting and a .NET SDK (Microsoft Agent Framework) against wider language reach or lab-specific runtimes elsewhere | | **Best for** | Architects and AI leads choosing an agent framework for a team with a fixed language, cloud, and model strategy | | **Where Microsoft Agent Framework leads** | .NET support, five built-in orchestration patterns, and Foundry hosting on Azure | | **What none of them replaces** | Shared enterprise context: definitions, lineage, ownership, and policies that every agent needs | --- ## What is Microsoft Agent Framework, and where did it come from? [Microsoft Agent Framework](https://atlan.com/know/ai-agent/microsoft/agent-framework/) (MAF) builds AI agents and multi-agent workflows in .NET, Python, and Go. According to the [Microsoft Learn overview](https://learn.microsoft.com/en-us/agent-framework/overview/) (2026), the Go SDK is in public preview, and Go does not yet have declarative agents, RAG, CodeAct, or functional workflows. The code lives in the [microsoft/agent-framework repository](https://github.com/microsoft/agent-framework) under an MIT license. The same overview calls MAF the direct successor to [Semantic Kernel](https://atlan.com/know/ai-agent/microsoft/semantic-kernel/) and [AutoGen](https://atlan.com/know/ai-agent/what-is-autogen/), built by the same teams. It keeps AutoGen's simple agent abstractions and Semantic Kernel's enterprise features, then adds graph-based workflows for explicit control over multi-agent execution paths. The other six frameworks fall into three groups. AWS Strands and Google ADK pair with a hyperscaler. The OpenAI Agents SDK and Claude Agent SDK come from frontier labs. [LangGraph](https://atlan.com/know/ai-agent/ai-agent-memory/what-is-langgraph/) and [CrewAI](https://atlan.com/know/ai-agent/what-is-crewai/) are independent, though [LangGraph's maker also sells LangSmith](https://docs.langchain.com/oss/python/langgraph/overview) for deployment. A separate question, covered in [agent harness vs. agent framework](https://atlan.com/know/ai-agent/agent-harness-vs-agent-framework/), is whether you want a library of building blocks or a packaged runtime. Several frameworks here offer both. MAF ships a Harness Agent, which [Microsoft Learn](https://learn.microsoft.com/en-us/agent-framework/concepts/harness) (2026) describes as an opinionated, batteries-included agent for research, coding, and other long-running work. For a broader view of what an [agent harness](https://atlan.com/know/what-is-an-agent-harness/) does, start there. The Harness Agent comes with a caveat for Go and Python teams. Microsoft's documentation says no packaged Go Harness exists today, and that Python's background agents, file access, and looping remain experimental. --- ## How does Microsoft Agent Framework compare with AWS Strands and Google ADK? All three frameworks come from hyperscalers, accept models from several providers, and offer a managed path on their own cloud. They split on languages, workflow design, and hosting. Teams weighing these two against [LangGraph and AutoGen](https://atlan.com/know/ai-agent/ai-agent-applications/google-adk-vs-langgraph-vs-autogen/) will find a longer treatment there, and the sibling page on [Google ADK](https://atlan.com/know/ai-agent/ai-agent-applications/what-is-google-adk/) covers that framework in depth. | Aspect | Microsoft Agent Framework | AWS Strands | Google ADK | |---|---|---|---| | Models | [Foundry, Azure OpenAI, OpenAI, Anthropic, Ollama, Amazon Bedrock, Google Gemini, and others](https://learn.microsoft.com/en-us/agent-framework/integrations/by-component/model-providers/), with feature coverage that varies by provider | [Anthropic, OpenAI, Google, Amazon Bedrock, Mistral, Ollama, and others](https://strandsagents.com/) | [Gemini directly, plus adapters for OpenAI, Ollama, and other models](https://adk.dev/) | | Languages | [.NET, Python, and Go (public preview)](https://learn.microsoft.com/en-us/agent-framework/overview/) | [TypeScript and Python](https://github.com/strands-agents/sdk-python) | [Python, TypeScript, Go, Java, and Kotlin](https://adk.dev/) | | Multi-agent design | [Sequential, Concurrent, Handoff, Group Chat, and Magentic](https://learn.microsoft.com/en-us/agent-framework/workflows/orchestrations/) orchestrations | [Agents as tools, Swarm, Graph, and Workflow patterns](https://strandsagents.com/), with [Graph](https://strandsagents.com/docs/user-guide/sdk/multi-agent/graph/) giving deterministic control | [Graph, dynamic, and collaborative workflows](https://adk.dev/workflows/) introduced in [ADK 2.0](https://adk.dev/2.0/) | | Protocols | MCP clients, [A2A](https://learn.microsoft.com/en-us/agent-framework/integrations/by-component/agent-services/a2a), and [AG-UI](https://learn.microsoft.com/en-us/agent-framework/integrations/by-component/ui/ag-ui/) | [MCP and A2A](https://strandsagents.com/docs/user-guide/sdk/multi-agent/agent-to-agent/) | [MCP through McpToolset](https://adk.dev/tools-custom/mcp-tools/), and [A2A (marked experimental)](https://adk.dev/a2a/) | | Managed hosting | [Foundry hosted agents](https://learn.microsoft.com/en-us/azure/foundry/agents/concepts/hosted-agents), the [Azure Functions durable extension](https://learn.microsoft.com/en-us/agent-framework/hosting/azure-functions), or your own compute | [Amazon Bedrock AgentCore Runtime](https://strandsagents.com/docs/user-guide/sdk/deploy/deploy_to_bedrock_agentcore/), Lambda, Fargate, App Runner, EKS, EC2, or Docker | [Agent Runtime, Cloud Run, GKE](https://adk.dev/), or a container on your own infrastructure | | License | [MIT](https://github.com/microsoft/agent-framework) | [Apache 2.0](https://github.com/strands-agents/sdk-python) | [Apache 2.0](https://github.com/google/adk-python) | Two details in that table change the decision. First, Foundry hosting does not lock you to MAF. Microsoft's hosted agents documentation says you can bring Agent Framework, LangGraph, Semantic Kernel, or custom code, and that hosted agents support Python and C#. Go agents cannot use that path today. Second, Strands also ships a packaged harness, which its [documentation](https://strandsagents.com/docs/user-guide/harness/) describes as a fully assembled agent harness with tuned defaults for tools, memory, and system prompts. If your infrastructure runs on Azure, MAF gives the shortest path to managed hosting. If you plan to run agents on your own infrastructure, the choice comes down to SDK maturity, language coverage, and workflow style. Teams that need .NET have one option here. Teams that need Java or Kotlin have one as well, and it is ADK. Teams hesitating between a hyperscaler framework and a [neutral context layer](https://atlan.com/know/ai-agent/context-layer/single-stack-lock-in-vs-neutral-context-layer/) should note that the two choices answer different questions. --- ## How does Microsoft Agent Framework compare with the OpenAI Agents SDK and Claude Agent SDK? The OpenAI Agents SDK and Claude Agent SDK come from companies that build frontier models, so they bundle runtime scaffolding that the hyperscaler frameworks leave to you. They also overlap with MAF in a useful way. MAF can use [OpenAI as a model provider](https://learn.microsoft.com/en-us/agent-framework/integrations/by-component/model-providers/), and its [Anthropic Claude integration](https://learn.microsoft.com/en-us/agent-framework/integrations/by-component/agent-services/anthropic-claude) (2026) wraps the Claude Agent SDK as a `ClaudeAgent`. That integration is a Python package, installed with the `--pre` flag. | Aspect | Microsoft Agent Framework | OpenAI Agents SDK | Claude Agent SDK | |---|---|---|---| | Models | Several providers, with [MCP, code interpreter, and file search support varying by provider](https://learn.microsoft.com/en-us/agent-framework/integrations/by-component/model-providers/) | [Built for OpenAI APIs, with support for 100+ other LLMs](https://github.com/openai/openai-agents-python) | [Claude models](https://code.claude.com/docs/en/agent-sdk/overview), reachable through Anthropic's API or [Amazon Bedrock and Google Cloud endpoints](https://code.claude.com/docs/en/agent-sdk/hosting) | | Languages | [.NET, Python, and Go (public preview)](https://learn.microsoft.com/en-us/agent-framework/overview/) | [Python](https://openai.github.io/openai-agents-python/) and [TypeScript](https://openai.github.io/openai-agents-js/) | [Python and TypeScript](https://code.claude.com/docs/en/agent-sdk/overview) | | Runtime scaffolding | [Harness Agent](https://learn.microsoft.com/en-us/agent-framework/concepts/harness) for long, multi-step work | [Sandbox agents](https://openai.github.io/openai-agents-python/sandbox_agents/) with persistent workspaces | Claude Code's agent loop, built-in tools, and sessions as a library, plus [Managed Agents](https://platform.claude.com/docs/en/managed-agents/overview) in beta | | Multi-agent design | Five orchestration patterns | [Agents as tools and handoffs](https://openai.github.io/openai-agents-python/multi_agent/) | [Subagents](https://code.claude.com/docs/en/agent-sdk/overview) for focused subtasks | | Safety controls | [Middleware and tool approval](https://learn.microsoft.com/en-us/agent-framework/workflows/human-in-the-loop) | [Guardrails](https://openai.github.io/openai-agents-python/) and human-in-the-loop mechanisms | [Permissions and hooks](https://code.claude.com/docs/en/agent-sdk/overview) | | Protocols | MCP clients, A2A, and AG-UI | [MCP through hosted tools, streamable HTTP, and stdio](https://openai.github.io/openai-agents-python/mcp/) | [MCP](https://code.claude.com/docs/en/agent-sdk/overview) for external tools and data sources | | Where it runs | Foundry hosted agents, Azure Functions, or your own compute | [A library you install](https://openai.github.io/openai-agents-python/) with `pip install openai-agents` | [Your own containers](https://code.claude.com/docs/en/agent-sdk/hosting) as a CLI subprocess per session, or Managed Agents | | License | [MIT](https://github.com/microsoft/agent-framework) | [MIT](https://github.com/openai/openai-agents-python) | [Anthropic's Commercial Terms of Service](https://github.com/anthropics/claude-agent-sdk-python) govern use | The license row deserves a second look. The Claude Agent SDK repository carries an MIT license file, but the README and the [SDK documentation](https://code.claude.com/docs/en/agent-sdk/overview) both state that Anthropic's Commercial Terms govern its use, with exceptions for components that carry their own license. Read both before you build a product on it. Hosting also differs in kind. The Agent SDK spawns a `claude` CLI subprocess per session, so Anthropic's hosting guide treats each running agent as a long-lived process tied to local state. [Claude Managed Agents](https://platform.claude.com/docs/en/managed-agents/overview) moves that loop onto Anthropic's infrastructure, but it is in beta and, per Anthropic, not eligible for Zero Data Retention or HIPAA BAA coverage today. Choose the OpenAI or Claude SDK when you commit to that lab's models, its workflow patterns, and its built-in harness. Choose MAF when .NET matters, or when you want to mix models and still keep one orchestration layer. Both lab SDKs speak MCP, which keeps your [context layer](https://atlan.com/know/what-is-context-layer/) portable across them. --- ## How does Microsoft Agent Framework compare with LangGraph and CrewAI? [LangGraph](https://atlan.com/know/ai-agent/ai-agent-memory/langchain-vs-langgraph/) and CrewAI do not belong to a cloud or a model lab. Both sell a managed platform on top of the open-source framework, and both document the same durable-execution ideas MAF does. | Aspect | Microsoft Agent Framework | LangGraph | CrewAI | |---|---|---|---| | Programming model | Agents plus [graph-based workflows](https://learn.microsoft.com/en-us/agent-framework/overview/) and five orchestration patterns | [State, nodes, and edges](https://docs.langchain.com/oss/python/langgraph/graph-api) in a low-level graph runtime | [Agents, Crews, and Flows](https://docs.crewai.com/), with [Crews](https://docs.crewai.com/en/concepts/crews) running sequential or hierarchical processes over tasks | | Languages | .NET, Python, and Go (public preview) | [Python](https://docs.langchain.com/oss/python/langgraph/overview) and [JavaScript](https://docs.langchain.com/oss/javascript/langgraph/overview) | [Python](https://docs.crewai.com/) | | Checkpoints | [Created at the end of each superstep](https://learn.microsoft.com/en-us/agent-framework/workflows/checkpoints), with in-memory, file, and Cosmos DB storage in Python | [Checkpointers](https://docs.langchain.com/oss/python/langgraph/persistence) save graph state snapshots, scoped to a thread | [Automatic checkpoints on events](https://docs.crewai.com/en/concepts/checkpointing), task completion by default, plus manual `state.checkpoint()` calls | | Human-in-the-loop | [`RequestPort`](https://learn.microsoft.com/en-us/agent-framework/workflows/human-in-the-loop) pauses a workflow for an external response | [Interrupts](https://docs.langchain.com/oss/python/langgraph/interrupts) save state and wait until you resume | [`@human_feedback`](https://docs.crewai.com/en/learn/human-feedback-in-flows) pauses a Flow for human review | | Protocols | MCP clients, A2A, and AG-UI | [MCP through LangChain](https://docs.langchain.com/oss/python/langchain/mcp), and [A2A in the LangSmith Agent Server](https://docs.langchain.com/langsmith/server-a2a) | [MCP](https://docs.crewai.com/en/mcp/overview), [A2A delegation](https://docs.crewai.com/en/learn/a2a-agent-delegation), and [A2UI served over AG-UI](https://docs.crewai.com/en/guides/frontend/a2ui) | | Managed platform | [Foundry hosted agents](https://learn.microsoft.com/en-us/azure/foundry/agents/concepts/hosted-agents) | [LangSmith](https://docs.langchain.com/oss/python/langgraph/overview) deployment | [CrewAI AMP](https://docs-platform.crewai.com/platform/en/introduction) for deploying, monitoring, and scaling crews | LangGraph also has a harness layer. [Deep Agents](https://docs.langchain.com/oss/python/deepagents/overview) sits on LangChain's core components and the LangGraph runtime, and adds filesystem tools, subagent delegation, and memory. That makes the three-way split in this section less clean than the table suggests, because MAF, Strands, and LangGraph all offer a packaged harness alongside the low-level framework. The table shows near parity on durable execution and human approval. All three document checkpointing and a way to pause for a person. The deciding factors are language and hosting. LangGraph and CrewAI run on any cloud with no Azure dependency. Neither lists a .NET SDK. Use MAF when you want Foundry to host your agents, or when .NET is a requirement. Use LangGraph when you want explicit graph control in Python or JavaScript, and CrewAI when the crew-and-flow model matches how your team already thinks. --- ## What do agents still need after you pick a framework? Whatever framework you choose, agents need the context of your organization: what your systems contain, what the data means, which assets to trust, and how policies apply. That context lives across dozens, sometimes hundreds, of systems, and each system's context stops at its own boundary. This gap is what [context engineering](https://atlan.com/know/what-is-context-engineering/) exists to close, and it is why teams compare [an agent context layer with RAG](https://atlan.com/know/ai-agent/agent-context-layer-vs-rag/) before they commit. [Atlan](https://atlan.com/) is the [Context Layer for AI](https://atlan.com/context-layer/). It spans enterprise systems and sits on the [**Context Lakehouse**](https://atlan.com/context-lakehouse/), where it holds lineage, quality, ownership, and policies. The [reference architecture](https://atlan.com/know/ai-agent/context-layer/context-layer-reference-architecture/) shows how those pieces fit together, and the [evaluation criteria](https://atlan.com/know/ai-agent/context-layer/context-layer-evaluation-criteria/) show what to test before you choose one. All seven frameworks document MCP support, so each one can reach the Atlan MCP server. According to the [Atlan documentation](https://docs.atlan.com/product/capabilities/atlan-ai/references/mcp-tools) (2026), that server exposes 39 tools across nine categories, including search and discovery, lineage and exploration, asset metadata, business glossary, and data quality. Every tool carries a Read, Write, or Admin access level, so you can scope what an agent may change. Teams that want implementation detail can start with the guides for [CrewAI](https://atlan.com/know/ai-agent/context-layer-in-crewai-implementation/) and [LangGraph](https://atlan.com/know/ai-agent-memory/context-layer-in-langgraph-implementation/). Atlan's constructs do the work behind that MCP connection: - The [Enterprise Data Graph](https://atlan.com/know/enterprise-data-graph/) is a queryable map of data assets, business definitions, policies, lineage, and quality rules. - The [Active Ontology](https://atlan.com/know/what-is-active-ontology/) holds technical and business concepts, glossary terms, domain definitions, and metrics. - [**Context Agents**](https://atlan.com/context-agents/) synthesize unstructured documentation and metadata into structured context for other agents. - [**Context Engineering Studio**](https://atlan.com/context-engineering-studio/) builds [Context Repos](https://atlan.com/know/ai-agent/context-repository-for-ai-agents/), runs context-based evaluations on your agents, and deploys context to them. The payoff is arithmetic. Wiring every enterprise system into every agent framework multiplies connections and review work each time a team adopts a new framework or [harness](https://atlan.com/know/context-layer-harness-engineering/). Connecting each system to the context layer once means a new framework needs one MCP connection, and [role-based access control](https://atlan.com/know/ai-agent/context-layer/context-layer-role-based-access-control/) applies in one place. The same logic holds across [clouds](https://atlan.com/know/ai-agent/context-layer/multi-cloud-context-layer/) and [models](https://atlan.com/know/ai-agent/context-layer/model-agnostic-context-layer/), and it also applies to the sibling comparison of [Databricks Omnigent and a neutral context layer](https://atlan.com/know/ai-agent/databricks/databricks-omnigent-vs-neutral-context-layer/). Teams that already run coding agents face the same gap, which [Claude Code and Codex users](https://atlan.com/know/ai-agent/context-layer/claude-code-codex-context-gap/) hit first. --- ## Which agent framework should you choose? Microsoft Agent Framework is the default choice when you run on Azure, need a .NET SDK, or want Foundry to host your agents and evaluations. Microsoft's own documentation also lets Foundry host LangGraph and custom code, so Foundry alone does not decide the question. AWS Strands and Google ADK play the same role on their clouds. Both accept models from other providers, and ADK lists the widest language range of any framework here. Pick the OpenAI Agents SDK or Claude Agent SDK when you commit to one lab. Pick LangGraph or CrewAI when you want a cloud-agnostic framework with its own managed platform. The [decision guide for enterprise agentic frameworks](https://atlan.com/know/ai-agent/how-to-choose-agentic-framework-enterprise/) walks through the questions in order, and the [comparison of AI agent harness tools](https://atlan.com/know/best-ai-agent-harness-tools-2026/) covers the runtime layer. For a similar cloud-versus-independent comparison on AWS, see [Bedrock Agents vs. LangGraph](https://atlan.com/know/ai-agent/ai-agent-applications/aws-bedrock-agents-vs-langgraph/), and for the Bedrock side of the data question, [Amazon Bedrock Knowledge Bases vs. an external data catalog](https://atlan.com/know/ai-agent/aws/amazon-bedrock-knowledge-bases-vs-external-data-catalog/). Interoperability narrows the stakes. [MCP](https://atlan.com/know/what-is-model-context-protocol/) and [A2A](https://atlan.com/know/google-a2a-protocol/) let agents from different frameworks share tools and talk to each other, as the overview of [agent interoperability protocols](https://atlan.com/know/agent-interoperability-protocols/) explains. Protocol choices still need [security review](https://atlan.com/know/ai-agent/ai-agent-governance/how-to-secure-multi-agent-systems-enterprise/) and a plan for [debugging multi-agent runs](https://atlan.com/know/ai-agent/debugging-multi-agent-systems/) once you run more than one framework in production. Teams building their own tool servers can follow the [MCP server implementation guide](https://atlan.com/know/mcp-server-implementation-guide/), and the sibling guide on [connecting an AI coding agent to a data warehouse over MCP](https://atlan.com/know/ai-agent/how-to-connect-ai-coding-agent-to-data-warehouse-mcp/) shows the same pattern from the coding-agent side. Which framework will your team still maintain in two years? That question deserves more weight than any feature row above. --- ## FAQs about Microsoft Agent Framework vs. other agent frameworks ### 1. What is Microsoft Agent Framework? Microsoft Agent Framework is an open-source framework for building agents and multi-agent workflows. It ships SDKs for .NET and Python, and the Go SDK is in public preview. It accepts models from several providers, including Microsoft Foundry, Azure OpenAI, OpenAI, Anthropic, and Ollama. ### 2. How is Microsoft Agent Framework related to Semantic Kernel and AutoGen? Microsoft describes Agent Framework as the direct successor to both, created by the same teams. It combines AutoGen's simple single-agent and multi-agent abstractions with Semantic Kernel's enterprise features, such as session-based state management, type safety, filters, and telemetry, and adds graph-based workflows for explicit control over multi-agent execution. ### 3. Can you run Microsoft Agent Framework agents outside Azure? Yes. The framework is MIT-licensed and model-agnostic, so you can run agents on your own compute with models from providers such as Anthropic, Google Gemini, Amazon Bedrock, and Ollama. Foundry hosted agents and Azure Functions give a managed path on Azure, while the durable hosting extension also supports bring-your-own compute. ### 4. Which agent and tool protocols does Microsoft Agent Framework support? It supports MCP for tool integration through MCP clients, A2A for talking to remote agents, and AG-UI for web-based agent applications with streaming, human-in-the-loop approvals, and shared state. AG-UI support varies by SDK, so check the language-specific documentation before you commit. ### 5. When should you choose Microsoft Agent Framework over other agent frameworks? Choose it when you need a .NET SDK or plan to run agents on Azure with Foundry hosting. It also fits teams that want five built-in orchestration patterns and a packaged harness agent. Choose another framework when your team works in Java, Kotlin, or TypeScript, or when you commit to one lab's models and runtime. ### 6. How can Microsoft Agent Framework agents connect to the Atlan MCP server? Microsoft Agent Framework includes MCP clients for tool integration, so an agent connects to the Atlan MCP server the way it connects to any MCP server. The agent then searches assets, traverses lineage, and reads governed context through Atlan's tools without a separate connection to each enterprise system. Other frameworks in this comparison document MCP support, so the pattern carries across them. --- ## Sources 1. Microsoft Agent Framework overview, Microsoft Learn, 2026. https://learn.microsoft.com/en-us/agent-framework/overview/ 2. microsoft/agent-framework, GitHub, 2026. https://github.com/microsoft/agent-framework 3. Workflow orchestrations in Agent Framework, Microsoft Learn, 2026. https://learn.microsoft.com/en-us/agent-framework/workflows/orchestrations/ 4. Agent Harness, Microsoft Learn, 2026. https://learn.microsoft.com/en-us/agent-framework/concepts/harness 5. Model providers, Microsoft Learn, 2026. https://learn.microsoft.com/en-us/agent-framework/integrations/by-component/model-providers/ 6. Workflow checkpoints, Microsoft Learn, 2026. https://learn.microsoft.com/en-us/agent-framework/workflows/checkpoints 7. Human-in-the-loop in workflows, Microsoft Learn, 2026. https://learn.microsoft.com/en-us/agent-framework/workflows/human-in-the-loop 8. A2A agent service, Microsoft Learn, 2026. https://learn.microsoft.com/en-us/agent-framework/integrations/by-component/agent-services/a2a 9. AG-UI integration with Agent Framework, Microsoft Learn, 2026. https://learn.microsoft.com/en-us/agent-framework/integrations/by-component/ui/ag-ui/ 10. Anthropic Claude agent service, Microsoft Learn, 2026. https://learn.microsoft.com/en-us/agent-framework/integrations/by-component/agent-services/anthropic-claude 11. Hosted agents in Foundry Agent Service, Microsoft Learn, 2026. https://learn.microsoft.com/en-us/azure/foundry/agents/concepts/hosted-agents 12. Durable extension for Agent Framework (Azure Functions), Microsoft Learn, 2026. https://learn.microsoft.com/en-us/agent-framework/hosting/azure-functions 13. Strands Agents, Strands Agents, 2026. https://strandsagents.com/ 14. strands-agents/sdk-python, GitHub, 2026. https://github.com/strands-agents/sdk-python 15. Strands harness, Strands Agents documentation, 2026. https://strandsagents.com/docs/user-guide/harness/ 16. Graph multi-agent pattern, Strands Agents documentation, 2026. https://strandsagents.com/docs/user-guide/sdk/multi-agent/graph/ 17. Agent-to-Agent protocol in Strands, Strands Agents documentation, 2026. https://strandsagents.com/docs/user-guide/sdk/multi-agent/agent-to-agent/ 18. Deploy Strands to Amazon Bedrock AgentCore, Strands Agents documentation, 2026. https://strandsagents.com/docs/user-guide/sdk/deploy/deploy_to_bedrock_agentcore/ 19. Agent Development Kit, Google, 2026. https://adk.dev/ 20. ADK workflows, Google, 2026. https://adk.dev/workflows/ 21. ADK 2.0, Google, 2026. https://adk.dev/2.0/ 22. ADK A2A protocol support, Google, 2026. https://adk.dev/a2a/ 23. MCP tools in ADK, Google, 2026. https://adk.dev/tools-custom/mcp-tools/ 24. google/adk-python, GitHub, 2026. https://github.com/google/adk-python 25. OpenAI Agents SDK, OpenAI, 2026. https://openai.github.io/openai-agents-python/ 26. openai/openai-agents-python, GitHub, 2026. https://github.com/openai/openai-agents-python 27. OpenAI Agents SDK for TypeScript, OpenAI, 2026. https://openai.github.io/openai-agents-js/ 28. Sandbox agents, OpenAI Agents SDK documentation, 2026. https://openai.github.io/openai-agents-python/sandbox_agents/ 29. Model Context Protocol in the OpenAI Agents SDK, OpenAI, 2026. https://openai.github.io/openai-agents-python/mcp/ 30. Agent orchestration, OpenAI Agents SDK documentation, 2026. https://openai.github.io/openai-agents-python/multi_agent/ 31. Agent SDK overview, Anthropic, 2026. https://code.claude.com/docs/en/agent-sdk/overview 32. Hosting the Agent SDK, Anthropic, 2026. https://code.claude.com/docs/en/agent-sdk/hosting 33. anthropics/claude-agent-sdk-python, GitHub, 2026. https://github.com/anthropics/claude-agent-sdk-python 34. Claude Managed Agents overview, Anthropic, 2026. https://platform.claude.com/docs/en/managed-agents/overview 35. LangGraph overview (Python), LangChain, 2026. https://docs.langchain.com/oss/python/langgraph/overview 36. LangGraph overview (JavaScript), LangChain, 2026. https://docs.langchain.com/oss/javascript/langgraph/overview 37. LangGraph Graph API, LangChain, 2026. https://docs.langchain.com/oss/python/langgraph/graph-api 38. LangGraph persistence, LangChain, 2026. https://docs.langchain.com/oss/python/langgraph/persistence 39. LangGraph interrupts, LangChain, 2026. https://docs.langchain.com/oss/python/langgraph/interrupts 40. Model Context Protocol in LangChain, LangChain, 2026. https://docs.langchain.com/oss/python/langchain/mcp 41. A2A in the LangSmith Agent Server, LangChain, 2026. https://docs.langchain.com/langsmith/server-a2a 42. Deep Agents overview, LangChain, 2026. https://docs.langchain.com/oss/python/deepagents/overview 43. CrewAI documentation, CrewAI, 2026. https://docs.crewai.com/ 44. Crews, CrewAI documentation, 2026. https://docs.crewai.com/en/concepts/crews 45. Checkpointing, CrewAI documentation, 2026. https://docs.crewai.com/en/concepts/checkpointing 46. Human feedback in Flows, CrewAI documentation, 2026. https://docs.crewai.com/en/learn/human-feedback-in-flows 47. MCP overview, CrewAI documentation, 2026. https://docs.crewai.com/en/mcp/overview 48. A2A agent delegation, CrewAI documentation, 2026. https://docs.crewai.com/en/learn/a2a-agent-delegation 49. A2UI in CrewAI, CrewAI documentation, 2026. https://docs.crewai.com/en/guides/frontend/a2ui 50. CrewAI AMP introduction, CrewAI, 2026. https://docs-platform.crewai.com/platform/en/introduction 51. Atlan MCP tools, Atlan Documentation, 2026. https://docs.atlan.com/product/capabilities/atlan-ai/references/mcp-tools