Repository navigation
Recoverability gate for AWS MCP calls: bounded live proof #4657
KingsMtn
started this conversation in
Show and tell
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Is this a useful integration direction for AWS MCP servers?
I’m building SMERC, a runtime gate that asks a different question from IAM or
--readonly: if this agent action fails, where will it execute, can the failure be contained, and how quickly can it be reversed?We ran a first bounded proof against the managed AWS MCP server:
aws-mcp/list_regionstarget, empty arguments, read-only scope, recoverability evidence, and a $0.00 cost declaration.ALLOWandshould_forward: true, with no missing-evidence failures.The evidence and its limitations are here:
The important limitation: this was a supervised two-step run. We have not yet technically forced the managed AWS MCP transport through SMERC, and we are not calling it production evidence.
For people working on the AWS MCP servers: which integration boundary would be most useful to validate next—a local MCP proxy, a managed gateway/AgentCore path, or correlation with native AWS audit evidence? I’d especially value criticism of the target binding, cost gate, and evidence receipt shape.
All reactions