Feature matrix · Last reviewed September 2026
Multi-agent framework feature matrix
Last reviewed: September 2026
Quick answer
LangGraph and Google ADK have the most Yes marks, 9 of 13 each, followed by BAND and Microsoft Agent Framework with 8. BAND is the only tool marked Yes on all three collaboration rows: cross-framework agents, humans as live participants and coding agents as teammates. The frameworks lead on open-source licensing, durable execution, built-in tracing and self-hosting; BAND is a hosted (SaaS) service with no documented self-hosting. Every tool supports MCP, and Google ADK leads on languages.
Collaboration and framework features matrix
| Feature | BAND | LangGraph | CrewAI | Microsoft Agent Framework | Google ADK (Agent Development Kit) | OpenAI Agents SDK | n8n | Agno |
|---|---|---|---|---|---|---|---|---|
| Python SDK | Yes | Yes | Yes | Yes | Yes | Yes | No / not documented: No SDK (code nodes only)No SDK (code nodes only) | Yes |
| TypeScript / JavaScript SDK | Yes | Yes | No / not documented | No / not documented: .NET and Go instead.NET and Go instead | Yes | Yes: @openai/agents@openai/agents | No / not documented: No SDK (code nodes only)No SDK (code nodes only) | No / not documented |
| .NET, Go, Java or Kotlin | No / not documented | No / not documented | No / not documented | Yes: .NET GA, Go preview.NET GA, Go preview | Yes: Go, Java, KotlinGo, Java, Kotlin | No / not documented | No / not documented | No / not documented |
| A2A protocol | Yes | Yes: LangSmith Agent Server, A2A v1.0LangSmith Agent Server, A2A v1.0 | Yes: native delegation, client and servernative delegation, client and server | Partial: open-source framework shows A2A in hosting samples; Microsoft Foundry Agent Service lists A2A v1.0 as generally availableopen-source framework shows A2A in hosting samples; Microsoft Foundry Agent Service lists A2A v1.0 as generally available | Yes: Google created A2AGoogle created A2A | No / not documented | No / not documented | Yes: AgentOS interfaceAgentOS interface |
| MCP | Yes | Yes: agents exposed at /mcpagents exposed at /mcp | Yes: mcps fieldmcps field | Yes: stdio, HTTP, WebSocket toolsstdio, HTTP, WebSocket tools | Yes: toolstools | Yes: MCP server toolsMCP server tools | Yes: client and server nodesclient and server nodes | Yes: tools and AgentOS as servertools and AgentOS as server |
| Connects agents built in other frameworks | Yes: native adapters and published integrations incl. LangGraph, CrewAI, Google ADK, Agno, Pydantic AI, Letta, Claude Agent SDK, Copilot Studio, Agentforcenative adapters and published integrations incl. LangGraph, CrewAI, Google ADK, Agno, Pydantic AI, Letta, Claude Agent SDK, Copilot Studio, Agentforce | Partial: via A2A endpointsvia A2A endpoints | Partial: A2A delegationA2A delegation | Partial: MCP tools; A2A samplesMCP tools; A2A samples | Partial: consume remote A2A agentsconsume remote A2A agents | No / not documented | Partial: MCP client/serverMCP client/server | Partial: A2AA2A |
| Humans as live participants in agent threads | Yes: rooms where people inspect, approve, override and reply in placerooms where people inspect, approve, override and reply in place | Partial: human-in-the-loop interruptshuman-in-the-loop interrupts | Partial: HITL checkpointsHITL checkpoints | Partial: orchestrations support HITLorchestrations support HITL | Partial: action confirmations and human inputaction confirmations and human input | Partial: needs_approval interruptionsneeds_approval interruptions | Partial: human approval for toolshuman approval for tools | Partial: confirmation and user inputconfirmation and user input |
| Coding agents (Claude Code, Codex, Copilot) as participants | Yes: BAND for desktop; integrations list Claude Code, Codex, GitHub Copilot, OpenCodeBAND for desktop; integrations list Claude Code, Codex, GitHub Copilot, OpenCode | No / not documented | No / not documented | No / not documented | No / not documented | No / not documented | No / not documented | No / not documented |
| Vendor-managed hosting | Yes: Hosted (SaaS); connects agents running on AWS, GCP or on-premHosted (SaaS); connects agents running on AWS, GCP or on-prem | Yes: LangSmith DeploymentLangSmith Deployment | Yes: CrewAI AMPCrewAI AMP | Yes: Microsoft Foundry hosted agentsMicrosoft Foundry hosted agents | Yes: Agent Runtime on Google CloudAgent Runtime on Google Cloud | No / not documented: library, you hostlibrary, you host | Yes: n8n Cloudn8n Cloud | Partial: hosted control plane on Pro; runtime runs in your cloudhosted control plane on Pro; runtime runs in your cloud |
| Self-hosting | No / not documented: No (not documented)No (not documented) | Yes: library; platform self-host on Enterpriselibrary; platform self-host on Enterprise | Yes | Yes | Yes | Yes | Yes: Community EditionCommunity Edition | Yes |
| Open-source license | No / not documented: commercial, not open sourcecommercial, not open source | Yes: MITMIT | Yes: MITMIT | Yes: MITMIT | Yes: Apache 2.0Apache 2.0 | Yes: MITMIT | No / not documented: fair-code Sustainable Use License, not OSI open sourcefair-code Sustainable Use License, not OSI open source | Yes: Apache-2.0Apache-2.0 |
| Durable execution / checkpointing | No / not documented: not documented; per-message delivery tracking insteadnot documented; per-message delivery tracking instead | Yes: checkpointers and durable executioncheckpointers and durable execution | Partial: Flow state persistenceFlow state persistence | Yes: workflow checkpointingworkflow checkpointing | Partial: session state servicessession state services | Partial: sessions; resumable run state for approvalssessions; resumable run state for approvals | Partial: saved executions; retry from failed nodesaved executions; retry from failed node | Partial: session storage in a databasesession storage in a database |
| Built-in tracing / observability | Partial: per-message lifecycle with attempt history; not a tracing toolper-message lifecycle with attempt history; not a tracing tool | Yes: LangSmith tracingLangSmith tracing | Yes: built-in tracingbuilt-in tracing | Yes: OpenTelemetryOpenTelemetry | Yes: dev UI traces; OpenTelemetrydev UI traces; OpenTelemetry | Yes: built-in tracingbuilt-in tracing | Partial: execution logs per runexecution logs per run | Partial: via observability integrationsvia observability integrations |
| Tally | 8 Yes, 1 Partial, 4 No | 9 Yes, 2 Partial, 2 No | 7 Yes, 3 Partial, 3 No | 8 Yes, 3 Partial, 2 No | 9 Yes, 3 Partial, 1 No | 6 Yes, 2 Partial, 5 No | 3 Yes, 4 Partial, 6 No | 5 Yes, 5 Partial, 3 No |
Yes = documented and generally available. Partial = available with limits, in preview, or only through samples or a separate package. No = not offered, or not documented in the vendor pages we reviewed in September 2026. Tell us if something changed: editors@multiagentcompare.com.
What does each feature mean?
- Python SDK / TypeScript SDK / .NET, Go, Java or Kotlin
- An official SDK maintained by the vendor. Code nodes inside a visual tool count as partial.
- A2A
- The Agent2Agent protocol, now a Linux Foundation project, for agent-to-agent calls across frameworks. v1.0.0 was released on March 12, 2026.
- MCP
- The Model Context Protocol, for connecting agents to tools and data. Latest spec revision 2026-07-28.
- Connects agents built in other frameworks
- Can agents that were not built with this tool take part in its multi-agent coordination? Partial means through a protocol call rather than as a full participant.
- Humans as live participants
- Can a person reply and redirect inside the same thread as the agents? Partial means approval or interrupt steps.
- Coding agents as participants
- Can Claude Code, Codex, GitHub Copilot or OpenCode sessions join and hand off work?
- Vendor-managed hosting / Self-hosting
- Whether the vendor runs it for you, and whether you can run it yourself.
- Open-source license
- Released under an OSI-approved open-source license such as MIT or Apache 2.0. Fair-code and source-available licenses count as No.
- Durable execution / checkpointing
- Can a run save its state and resume after a failure or a pause? Partial means session or execution storage without full workflow checkpointing.
- Built-in tracing / observability
- Does the tool record traces of agent runs, directly or through a first-party service? Partial means logs or third-party integrations.
How do coordination models and pricing compare?
| Tool | Coordination model | Entry price | License | Source |
|---|---|---|---|---|
| BAND | @mention-based routing in shared rooms (direct, group, task-scoped); per-agent, per-message lifecycle with attempt history; loop prevention via mandatory mentions and per-room limits; cross-agent context via Memories. | Free $0 (single user, non-commercial, up to 20 remote agents, 250 active rooms, 2-week retention); Pro $17.99/mo (single user, 40 remote + 40 native agents, 500 rooms, 40 participants per chat, data export); Enterprise custom (by quote; required for team or commercial use). | Commercial | BAND platform |
| LangGraph | Explicit state graph; documented multi-agent patterns are subagents, handoffs, skills, router and custom workflow. | OSS free; LangSmith Developer $0 (1 seat, 5k base traces/mo); Plus $39/seat/mo (10k base traces, 1 free Serverless Small deployment); Enterprise custom. | MIT | LangGraph repo |
| CrewAI | Crews of role-based agents run as a Sequential or Hierarchical process (hierarchical automatically assigns a manager); Flows add event-driven control with state. | OSS free; AMP Basic free (50 workflow executions/mo); Enterprise custom. | MIT | CrewAI repo |
| Microsoft Agent Framework | Typed, graph-based Workflows plus built-in orchestrations: Sequential, Concurrent, Handoff, Group Chat, Magentic; all support human-in-the-loop; checkpointing. | Framework free; Foundry and model usage billed by Azure. | MIT | Agent Framework repo |
| Google ADK (Agent Development Kit) | Sequential, Parallel and Loop workflow agents; graph-based workflows (new in 2.0); delegation via a Task API. | Framework free; runtime and models billed by Google Cloud. | Apache 2.0 | ADK docs |
| OpenAI Agents SDK | Handoffs between agents and agents-as-tools (manager style); guardrails, sessions, tracing built in; human approval of tool calls via needs_approval interruptions. | SDK free; pay model API usage. | MIT | OpenAI Agents SDK (Python) |
| n8n | Trigger- and event-driven workflows with an agent loop; agents can delegate to other agents (sub-agents) and run on a schedule; human approval before an AI Agent runs specific tools. | EUR, billed annually: Starter €20/mo (2,500 executions), Pro €50/mo (10,000), Business €667/mo (40,000, self-host option), Enterprise custom; Community Edition self-hosted free; one agent turn counts as one execution. | Sustainable Use License (fair-code, not OSI) + n8n Enterprise License | n8n pricing |
| Agno | Leader-and-members Teams with four modes (Coordinate, Route, Broadcast, Tasks) plus Workflows; HITL confirmations and user input. | Free $0; Pro $150/mo (1 live control-plane connection, 3 seats); Enterprise custom. | Apache-2.0 | Agno teams |
Official SDK languages
| Tool | Official SDK languages | Languages |
|---|---|---|
| BAND | 2 | Python, TypeScript |
| LangGraph | 2 | Python, JS/TS |
| CrewAI | 1 | Python |
| Microsoft Agent Framework | 3 | Python, .NET, Go preview |
| Google ADK | 5 | Python, TS/JS, Go, Java, Kotlin |
| OpenAI Agents SDK | 2 | Python, TS/JS |
| n8n | 1 | visual builder, JS/Python code nodes |
| Agno | 1 | Python |
Protocol support
| Tool | A2A score | MCP score |
|---|---|---|
| BAND | 1 | 1 |
| LangGraph | 1 | 1 |
| CrewAI | 1 | 1 |
| Microsoft Agent Framework | 0.5 | 1 |
| Google ADK | 1 | 1 |
| OpenAI Agents SDK | 0 | 1 |
| n8n | 0 | 1 |
| Agno | 1 | 1 |
What do people ask about the feature matrix?
Why is every tool marked Yes for MCP?
Because every one of them documents MCP support. That is why we do not treat MCP as a differentiator.
Why does BAND show No for self-hosting?
BAND is a hosted (SaaS) platform. band.ai says agents connected to BAND can run on different clouds (AWS, GCP, on-prem), but it does not document self-hosting BAND itself.
Why is AutoGen not in the matrix?
AutoGen is in maintenance mode. We compare it on two vs pages and recommend Microsoft Agent Framework for new work.
Where do these facts come from?
Reviewed Sep 2026
- BAND platform
- BAND pricing
- BAND integrations
- BAND docs
- LangGraph repo
- LangChain multi-agent docs
- LangSmith A2A
- LangSmith MCP
- LangChain pricing
- CrewAI repo
- CrewAI A2A docs
- CrewAI MCP docs
- CrewAI pricing
- Agent Framework repo
- Agent Framework releases
- Agent Framework overview
- Orchestrations overview
- Migrate from AutoGen
- Foundry Agent Service
- ADK docs
- adk-python
- A2A protocol
- OpenAI Agents SDK (Python)
- Human-in-the-loop
- Agents SDK (JS)
- AgentKit
- n8n pricing
- n8n agents docs
- n8n human-in-the-loop for tools
- n8n repo
- Agno teams
- Agno A2A
- Agno pricing