Hi Guanming,

I’ve read the drafts and think they are very interesting and exhibit a 
promising  and least resistance path to realize the vision of intent 
networking. While there are a lot of things to talk about, here I just raise 
one question for discussion.

The document doesn't cover the local server scenario. An alternative 
architecture is to keep the servers in controller with the network management 
application (i.e., local to the clients) and some resources/tools in network 
devices so the resources and tools can be accessed through the existing 
controller-device interface and protocols (e.g., NETCONF). While this may 
appear less appealing than the proposed architecture in the drafts, it's good 
to understand the pros and cons of each architecture in order to make the 
optimal decision. The alternative architecture at least shows advantages on the 
aspects of security, scalability, and development effort.

What are the compelling reasons to promote the proposed architecture only? 
Perhaps the mix of the two is better? Look forward to seeing your thoughts.

Best regards,
Haoyu


From: zengguanming <[email protected]>
Sent: Friday, November 7, 2025 2:18 AM
To: [email protected]
Cc: shangxiaotong <[email protected]>; Liubing (Leo) 
<[email protected]>
Subject: [rtgwg] MCP/A2A for Network Devices – New Draft Cluster and Call for 
Feedback / Co-Author

Dear Colleagues,

We would like to introduce a freshly-published cluster of five Internet-Drafts 
that explore the use of the MCP/A2A for AI-driven network devices management, 
and to ask for your advices, comments and collaboration.

1. A LOGIC-CHAIN IN FIVE DRAFTS
Regarding the aspect of LLM driven network operations and management, these 
five drafts comprehensively analyze the logic chain: scenario, use case and 
requirement analysis -> limitations of traditional protocols -> gaps in new 
protocols such as MCP/A2A -> protocol extensions:
(1) Define new application scenarios and corresponding use cases for network 
operation and management driven by LLMs, and identify the capabilities that 
protocols need to provide. 
(draft-zm-rtgwg-mcp-network-measurement-01<https://datatracker.ietf.org/doc/draft-zm-rtgwg-mcp-network-measurement/>,
 
draft-zm-rtgwg-mcp-troubleshooting-01<https://datatracker.ietf.org/doc/draft-zm-rtgwg-mcp-troubleshooting/>)
(2) Analyze the shortcomings of traditional network protocols (e.g., NETCONF) 
in implementing new application scenarios for network operation and management, 
and evaluate the feasibility of using AI-related protocols such as MCP and A2A. 
(draft-zeng-opsawg-applicability-mcp-a2a-00<https://datatracker.ietf.org/doc/draft-zeng-opsawg-applicability-mcp-a2a/>)
(3) Currently, MCP and A2A protocols are primarily designed for non-network 
management scenarios. This draft analyzes the gaps in protocol evolution when 
applied to network operation and management scenarios. 
(draft-zeng-opsawg-llm-netconf-gap-00<https://datatracker.ietf.org/doc/draft-zeng-opsawg-llm-netconf-gap/>)
(4) Based on the identified gaps in protocol evolution, attempts are made to 
extend the MCP protocol. 
(draft-zw-opsawg-mcp-network-mgmt-00<https://datatracker.ietf.org/doc/draft-zw-opsawg-mcp-network-mgmt/>)

2. ADVICES/ COMMENTS / COLLABORATION
Since the work is a little complex, we are sincerely need you help:
(1) MCP/A2A live in the application/AI space, but our use cases are deeply 
routing & operations oriented. What IETF areas and WGs do you think are 
appropriate for these drafts? How to coordinate these work if they are in 
different areas and WGs.
(2) Solicit your comments on the solution for MCP/A2A applied in network 
management and operation.
(3) Solicit your comments on these drafts.

These effort is still in early days — we are actively seeking broad input, 
additional use-cases, and co-authors to help shape the direction. Wish you 
could help and join the work.

Best regards,
Guanming Zeng
Huawei
E-mail: [email protected]<mailto:[email protected]>

_______________________________________________
rtgwg mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to