Repository navigation
Agent discovery extension (agents-wg#22): server conformance scenarios #576
Description
Agent discovery (agents-wg#22) merged in the Agents IG repo on 2026-10-20, after the WG vote.
I'd like to add server scenarios for it. #556 has a draft I put up before the vote so there was a diff to talk about.
What's in scope
Two server scenarios, graded off a requirement yaml that quotes the merged proposal at 137d88bd.
| Scenario | Grades |
|---|---|
sep-9998-agents-discovery |
the io.modelcontextprotocol/agents capability, agents/list cards and paging, the flat agents/get result, tools matching tools/list, the agent_not_found and invalid_agent_definition reasons, a hidden agent looking like an unknown one |
sep-9998-agents-list-changed |
notifications/agents/list_changed over subscriptions/listen with agentsListChanged |
49 rows in the yaml, 31 graded. The other 18 are excluded with a reason each: host obligations, rules that compare two authorization contexts (a run holds one credential), and step-up.
Things I'd like a read on
- The proposal has no SEP number, because it lives in the IG repo and not as a spec PR. The traceability filters only take numeric ids, so I used
sep-9998as a placeholder (9999 was the Events one before SEP-3415). Happy to rename by script when it gets a number. - The extension isn't on the spec timeline, so the scenarios only run with
--force, same as feat(sep-3415): Conformance suite for triggers and events #521. - The hidden-agent, broken-agent and
list_changedrows need two optional diagnostic tools on the server, the same pattern feat(sep-3415): Conformance suite for triggers and events #521 uses. Without them those rows report untestable. - One rule is coming that I haven't graded yet. Luca said on Discord that an undeclared client should get
-32021, and that's a pending follow-up to the proposal. I'll add a row once the text lands.
Implementations
mcpkit (Go) passes 28/28 and 4/4. Madhavi is working on the Python prototype, and I'd like to run it there next.
Activity