Skip to content

everything: add focused annotation-presence registration tests #5156

Closed

Description

@jw-ond

Which server?

everything (@modelcontextprotocol/server-everything)

Protocol era

Both, or not era-specific. The local verification below covers SDK 1 only; the request should be evaluated against the current v2/main registration paths.

MCP client

Automated Vitest tests. No external or LLM client was used for this test-only change.

The problem

Following the current issue-only contribution policy, I am converting the suggestion in #4529 into a registration-test coverage request, without attaching a patch.

The existing surface.test.ts already snapshots advertised annotations. The smaller gap is that the registration suite does not explicitly check the presence and boolean type of all four hints across every supported standard, capability-dependent, and task registration path. A focused assertion would give a more local failure when a refactor drops metadata at registration.

This is not a report that current tools lack annotations. Annotation presence also does not establish tool safety or prove that a hint accurately describes its behavior.

Solution you have in mind (optional)

Add registration-level regression checks for readOnlyHint, destructiveHint, idempotentHint, and openWorldHint, with the capabilities needed to exercise conditional/task paths enabled. Preserve the existing surface snapshots, capability tests, hint values, and validation gates. No runtime changes or provider-specific integration.

For current SDK 2 development, derive expectations from its modern/legacy registration paths rather than copying the SDK 1 counts below.

Which MCP feature does it demonstrate? (optional)

Existing MCP tool-annotation metadata and its registration paths.

Alternatives or workarounds you have tried

Relying on the existing wire-level snapshots alone remains a reasonable alternative if maintainers consider the additional registration check redundant.

I searched existing issues. #3401 concerns missing annotations in an older version, while #4854 tracks the completed broader coverage effort. This request does not repeat the older zero-annotations claim.

Already built it locally?

Yes, a test-only prototype against main at bbb701e7ac1e57f3edd1c2836b94c5457eaf7b9c (Everything 1.0.0, SDK 1). It preserves the ten existing registration test cases and adds two checks covering the 12 standard, six conditional, and one task registration in that snapshot. No production source changed. Those counts are not asserted for SDK 2.

Verified on 2026-10-15 with Node 22.18.0, npm 10.9.3 and Vitest 4.1.11:

  • npm test --workspace @modelcontextprotocol/server-everything -- __tests__/registrations.test.ts: 12 passed.
  • npm run validate --workspace @modelcontextprotocol/server-everything: formatting, lint, type checking, build, and 260 tests passed.
  • npm run coverage --workspace @modelcontextprotocol/server-everything: 260 tests passed and per-file coverage thresholds cleared.
  • git diff --check: passed.

These are scoped test/build results, not a claim that the repository-wide submission gate passed. The old PR's original commit predates the current DCO workflow. I have not rewritten that history or pushed the local repair.

Original repair prompt (exact quotation except two redacted local paths)

These historical instructions explain the local reproduction; this issue follows the current contribution policy instead of requesting a PR merge.

You own ONLY the existing clone [LOCAL_SERVERS_CLONE] for repairing #4529 (author jw-ond). User authorized beginning the upstream repair plan. Parent handles promptfoo, CrewAI, Langfuse. Do not edit [UNRELATED_LOCAL_PROJECT] or other clones. No new worktree. First inspect git status/branch/remotes, applicable AGENTS/CONTRIBUTING, and live PR head. If unexpected conflicting user changes exist, stop/report. Fetch current base and fork PR head, preserve all existing commits; prefer a merge from upstream main rather than history rewriting/force. Reproduce the conflict, resolve preserving intended tool-annotation registration regression coverage and current upstream behavior. Use apply_patch for manual edits, run appropriate focused tests/build plus git diff --check. This is restoring compatibility, no feature expansion and no test weakening. You may make a scoped normal commit on the verified PR feature branch; DO NOT PUSH or POST until parent reviews. Return old/new SHAs, exact changed file set, test commands/results, and any blockers. Work directly in that clone and preserve unrelated files. Do not disable hooks/security gates or use --no-verify.

Before you submit

  • I searched existing issues and this is not a duplicate.
  • This is a request for this repository or one of its servers, not a new server, the MCP specification or an SDK.

Activity

  1. added
    server-everythingReference implementation for the Everything MCP server - src/everything
    choreMaintenance: deps, build tooling, CI, cleanup — no user-facing behavior change
    on Oct 19, 2026
  2. cliffhall commented on Oct 19, 2026

    @cliffhall
    Member

    Triage: Priority Low (total 3)

    • Severity 1: A registration-level check that every tool sets all four annotation hints is a reasonable test-hygiene addition alongside the existing surface snapshot.
    • Urgency 2: wanted eventually
    • Bonuses: none

    Board: Servers V2 (#43), Status Incoming, awaiting maintainer review (no milestone yet). Part of the backlog triage in #4875.

  3. added this to the v2.3.0 milestone on Oct 25, 2026
  4. self-assigned this
    on Oct 28, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

choreMaintenance: deps, build tooling, CI, cleanup — no user-facing behavior changeserver-everythingReference implementation for the Everything MCP server - src/everythingv2

Type

No type

Projects

No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions