Show and tell: a read-only MCP for evaluating a live commerce task before any write #7838
aida-mikechen
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.
Full disclosure: I operate Plushwipes / Breezily24, the merchant behind this experiment. This is not a native Microsoft Agent Framework integration or an endorsement claim.
I wanted to make one real-world agent task inspectable before an operator authorizes any binding action. The public endpoint exposes a stateless, read-only MCP with three tools:
evaluate_opportunity— returns the live program status, terms, economics, and activation requirements;get_public_resource— returns the current public policy, terms, product, claims, and service resources;prepare_activation— produces a dry-run activation plan only. It does not register, accept terms, submit an identity, or create a credential.The underlying task is a live DTC performance program. The owner receives 10% of verified attributed net product sales, with $0 upfront cost, no required purchase or inventory, and no earnings guarantee. Registration and terms acceptance remain explicit human-approved writes.
For this sprint, a signup is not considered activation. An activated agent must register and then produce at least one legitimate external referral click through its own tracking flow.
Source-specific machine-readable skill:
https://earn.plushwipes.com/s/src_22a67576f56bc2ecacc1e695/skill.md
Read-only MCP endpoint:
https://plushwipes-agent-revenue-mcp.mike-01a.workers.dev/mcp
Source and documentation:
https://github.com/aida-mikechen/plushwipes-agent-revenue
The source-specific skill URL measures this discussion as a channel; it is not a participant referral link and opening it merely to test attribution would contaminate the measurement.
For Agent Framework builders: how would you divide this into evaluator, policy reviewer, campaign planner, and performance-monitor roles while keeping registration, publishing, messaging, and spend behind explicit owner approval? I am especially interested in the conditions that should make the workflow decline the task.
All reactions