Will this framework support MCP and A2A?

[hsuyuming]

Hi Team:

I find out there is a Teams AI Library v2 SDK nodejs version which support MCP and A2A. Do we have any plan to implement the same feature within this python version?

[1] https://learn.microsoft.com/en-us/microsoftteams/platform/teams-ai-library/in-depth-guides/ai/mcp/mcp-client

[stevkan]

Just clarifying. By A2A, do you mean agent to agent? To answer your question...

MCP - There is no current plan to support MCP, at this time.
A2A (assuming 'yes' to the above question) - The current SDK supports skill to skill, but these are Bot Framework bots. It does not natively support connecting to other bots or agents outside of Bot Framework, at this time.

[MattB-msft]

@hsuyuming

We are not adding new features to the Botbuilder-python any longer.. work has moved to the AgentSDK here: https://github.com/microsoft/agents-for-python/. You can find out more about the agent sdk here: https://github.com/microsoft/agents/

As part of this shift, we are moving the Agent SDK away from shipping its own Orchestrator / planner system, as we did in BotFramework.

What that means is that support for MCP will come down to which Orchestrator you chose to use and what level of support it provides for interfacing with MCP hosted tools. for example, Semantic Kernal for Python is working on support for MCP, and as you can see here: https://devblogs.microsoft.com/semantic-kernel/integrating-model-context-protocol-tools-with-semantic-kernel-a-step-by-step-guide/, they have presented a walk though for C#. OpenAI has also presented a capability to interface with MCP as part of their Orchestrator.

Insofar as A2A, what level of support and how it will surface is still under discussion and review. There are a few ways it could be done and surfaced.

The TeamsAI library provides its own Orchestrator, the documentation you're siting above is referring to use of that orchestrator.

[mathewnxwang]

Hey @MattB-msft what's the difference between the AgentSDK and the Teams AI Library? Which would you recommend using for building a chatbot inside Teams?

[hsuyuming]

Hi @stevkan @MattB-msft :

1. yes, for A2A i mean agent to agent.
2. If i understand correctly, botbuilder-python is build by traditional chat bot interface (e.g: dialogue, rule base...)? What does "we are not adding new features" mean? does this mean this package will be deprecated soon?
3. When i look at "https://github.com/microsoft/teams.ts", it looks like this typescript adk, can support user use mcp and a2a in any way? For example, i can use any model with MCPPlugin. If i understand wrong, please correct me.
4. This is our problem statement, we would like to build a agentic teams chatbot, we already have buildin mcp server, different agent services (e.g: build by crewai or GoogleADK), Which python package can allow us to develop a team chatbot UI, and also able to integrate with those MCP/Agent service?

[MattB-msft]

@mathewnxwang
The AgentSDK is designed to support agents developed to server any channel(client) using any orchestrator stack and is focused on providing core services to enable the hosting, Conversation management, User Authorization / Access token management, and Channel adapter capabilities.
Teams, as a channel, as some very specific features that are unique to teams only. Things like Message Extensions, Teams Channels, Meetings and such. In the Agent SDK these are accessed via the Teams Extensions, which bring across many of the features of TeamsAI lib and AI orchestration.

Teams AI Library is intended to be used only for Teams Clients. In support of that. all of its functions are centered around building an Agent Only for Teams, using the teams' features. Teams AI Lib also brings its own take on an Orchestrator which is a rough equivalent of an earlier version of Semantic Kernel, where that Orchestrator is predisposed to understand Graph and Teams' extended features.

@hsuyuming
re 2. there are some announcements we will be making at Build Next week.
re 3. Teams AI Lib is its own "thing" and we are working closely with that team to enabled access to the TeamsAI Lib AI system as an extension in the Agents SDK, this is partly complete in the .net version now.
You also have the option to use any orchestrator (or multiple if you want) to get at the various features you're looking for. Semantic Kernel, Autogen, OpenAI SDK and such all support MCP Tools.
re 4. You have 3 options right now for this..

• Start in the preview on AgentSDK Python version and utilize Semantic Kernel python to interface with your MCP tools.
• Start in TeamsAI v1 Python, replace the AI Lib with the Semantic Kernel / OpenAI SDK,
• Start in BotFramework Python and stick with using the 'core' activity, state and authorization systems, bringing in Semantic Kernel, OpenAI's or Anthropic's Orchestrators.

[hsuyuming]

Hi @MattB-msft :
Thank you for your input, Can we open this issue after Build? If you can provide any session link for re2. That's will great. Thank you very much