Why must the plugin use MCP instead of just using the AWS CLI?

[xiaosier]

Couldn't you just use AWS CLI to perform all the operations? Installing MCP requires starting an additional process locally. Why don't you choose to start an MCP and perform the operations through its tools instead of using CLI to perform all the operations in the skill directly?

[scottschreckengaust]

Can you confirm which plugin (in particular) so this can be answered or clarify if this is a general comment to explore? Thanks!

[xiaosier]

Thank you for your reply.

For example, in https://github.com/awslabs/agent-plugins/blob/main/plugins/aws-serverless/skills/api-gateway/references/deployment.md#manual-rollback-via-deployment-history-rest-api, a CLI call example is declared in the skill, but MCP is also declared in this project, such as in https://github.com/awslabs/agent-plugins/blob/main/plugins/aws-serverless/.mcp.json. When using this plugin, the local AWS CLI must be installed and logged in, because the MCP connection also reads the local credentials.

The question is, since the AWS CLI is already logged in, wouldn't the agent just use the local CLI to execute? Why does AWS still recommend prioritizing the use of MCP? Is there any special consideration?

[bfreiberg]

Thanks for the additional context @xiaosier , my thoughts:

since the AWS CLI is already logged in, wouldn't the agent just use the local CLI to execute?

I get your point for some of the actions. On the other hand, the Serverless MCP server has tools which do not directly map to AWS API actions such as deploy_webapp or get_lambda_event_schemas

Why does AWS still recommend prioritizing the use of MCP?

I don't see any indication that one method is recommended over another. Where did you perceive this? And nothing is preventing the agent to use the local CLI if it has enough confidence.