You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Is there an opportunity to contribute a solution to A2A Support in Agents SDK for Python? If so, is there a reference implementation or preferred approach that should be followed?
Is there an opportunity to contribute a solution to A2A Support in Agents SDK for Python? If so, is there a reference implementation or preferred approach that should be followed?
We're still planning how we would like to support A2A across the Agents SDKs. If you have any suggestions for how that implementation should look, we would love to get your input.
We could begin by mapping the existing Agents SDK abstractions to their A2A counterparts where there is a suitable match, for example defining how RequestContext maps to TurnContext.
Where direct mapping does not exist on either the Agents SDK or A2A side, additional bridging implementation may be required. For A2A components which do not currently have an equivalent in the existing Agents SDK architecture, such as AgentCard, an integration into the Agents SDK could be proposed.
If possible, I would love the opportunity to work on this implementation while confirming architectural decisions with you and your team.
We are leaning towards this pattern seen in the .NET code there, but we will gladly take additional feedback. The Python port will be slightly different (just as how the Python MSTeams extensions differs from the .NET MSTeams extension), but the overall design with the A2A Activity, TurnContext, and adapter would stay if we continue down this route.
The .NET architecture should translate well to the Python version, including an A2A Activity conversion layer, an A2A-specific TurnContext wrapper, and an adapter for converting A2A requests and responses through the existing Agents SDK pipeline.
For the Python implementation, the patterns in microsoft-agents-hosting-msteams could be drawn from rather than directly following the .NET registration APIs, allowing for a more Python-specific implementation. Thank you for the direction on this.
I could start with the Activity, TurnContext, and adapter, then move on to the AgentCard and endpoint integration. If changes are needed at any point, I’ll be happy to incorporate them.
The .NET architecture should translate well to the Python version, including an A2A Activity conversion layer, an A2A-specific TurnContext wrapper, and an adapter for converting A2A requests and responses through the existing Agents SDK pipeline.
For the Python implementation, the patterns in microsoft-agents-hosting-msteams could be drawn from rather than directly following the .NET registration APIs, allowing for a more Python-specific implementation. Thank you for the direction on this.
I could start with the Activity, TurnContext, and adapter, then move on to the AgentCard and endpoint integration. If changes are needed at any point, I’ll be happy to incorporate them.
I am open to taking advice and feedback on the design now and going forward, but I will be taking charge of the initial implementation.
Activity
[-]A2A Support in Agents SDK[/-][+]A2A Support in Agents SDK for Python[/+]seaic-mac-murchadha commentedon Aug 21, 2026
Hi Sarah Critchley (@sarahcritchley) ,
Is there an opportunity to contribute a solution to A2A Support in Agents SDK for Python? If so, is there a reference implementation or preferred approach that should be followed?
rodrigobr-msft commentedon Sep 2, 2026
We're still planning how we would like to support A2A across the Agents SDKs. If you have any suggestions for how that implementation should look, we would love to get your input.
seaic-mac-murchadha commentedon Sep 6, 2026
Hi Rodrigo Brandão (@rodrigobr-msft),
We could begin by mapping the existing Agents SDK abstractions to their A2A counterparts where there is a suitable match, for example defining how
RequestContextmaps toTurnContext.Where direct mapping does not exist on either the Agents SDK or A2A side, additional bridging implementation may be required. For A2A components which do not currently have an equivalent in the existing Agents SDK architecture, such as
AgentCard, an integration into the Agents SDK could be proposed.If possible, I would love the opportunity to work on this implementation while confirming architectural decisions with you and your team.
rodrigobr-msft commentedon Sep 10, 2026
In .NET, the A2A extension actually made its way into Preview: https://github.com/microsoft/Agents-for-net/tree/main/src/libraries/Extensions/Microsoft.Agents.Extensions.A2A
We are leaning towards this pattern seen in the .NET code there, but we will gladly take additional feedback. The Python port will be slightly different (just as how the Python MSTeams extensions differs from the .NET MSTeams extension), but the overall design with the A2A Activity, TurnContext, and adapter would stay if we continue down this route.
seaic-mac-murchadha commentedon Sep 11, 2026
Thanks Rodrigo Brandão (@rodrigobr-msft),
The .NET architecture should translate well to the Python version, including an A2A Activity conversion layer, an A2A-specific
TurnContextwrapper, and an adapter for converting A2A requests and responses through the existing Agents SDK pipeline.For the Python implementation, the patterns in
microsoft-agents-hosting-msteamscould be drawn from rather than directly following the .NET registration APIs, allowing for a more Python-specific implementation. Thank you for the direction on this.I could start with the Activity,
TurnContext, and adapter, then move on to theAgentCardand endpoint integration. If changes are needed at any point, I’ll be happy to incorporate them.rodrigobr-msft commentedon Sep 11, 2026
I am open to taking advice and feedback on the design now and going forward, but I will be taking charge of the initial implementation.