[JanGoebel]

Summary

Review findings on the CI tooling. The main one: with -CleanStart CI force-closed every running LabVIEW on the runner whenever a VIPM Desktop was open. The runner is the development machine, so that included the maintainer's own LabVIEW and anything unsaved in it. Now CI closes only the test and comparison LabVIEWs it chose for the job, and none with unsaved changes. Alongside: the workflow templates get least-privilege tokens, pinned actions and a release guard, the report scripts stop ignoring git and gh failures, and LabVIEW and Caraya text is escaped in the reports. No VI is changed.

	Before	After
VIPM Desktop open at job start	every running LabVIEW on the machine force-closed	only the job's test and comparison LabVIEWs, and none with a * (unsaved changes) in a window title unless CI_DEDICATED_MACHINE=1
LabVIEW on the wrong VI Server port (-Fix)	force-closed	same rule: kept when it has unsaved changes, LabVIEW.ini then not written while it runs
Release from a v* tag	any tagged commit, no approval, whole workflow with contents: write	only commits on main, job in the release environment, write access only for that job
Pull request checkouts	token persisted in .git/config while the PR's code runs	persist-credentials: false; only the publishing step gets GH_TOKEN
Setup or package install fails	protocol tests and VI diff run anyway with empty LABVIEW_*	they are skipped and reported as "not run"; the last step says why
git or gh call fails in the report scripts	mostly ignored (a failed git diff looked like "no VI changed")	the step fails with the git/gh error
Report comment lookup	first 100 comments, any comment with the marker overwritten	all pages, only the comment with the marker written by the same user
|, <, >, @ in labels and LabVIEW texts	broke table rows, vanished as tags, mentioned users	shown as written
Caraya assert level	deepest indent: one deeper line hid all tests; assert labels cut at " - "	by the report's structure; full labels

Changes

All files are PowerShell scripts, the two workflow templates and docs. .github/workflows/ is not touched.

• tools/ci/LabVIEWCi.psm1 (modified)
  • Select-LabVIEWToStop decides which running LabVIEWs may be closed, with no side effects: only paths the job uses, none with * in a window title unless CI_DEDICATED_MACHINE=1. Get-ProcessWindowTitle reads the titles of all visible top-level windows of a process (EnumWindows), Get-RunningLabVIEW and Stop-CiLabVIEW apply it, Test-CiDedicatedMachine reads the variable.
  • Stop-LeftoverVipm and Resolve-ViServerPort -Fix use that rule; their doc comments say that they force-close.
  • Invoke-NativeCommand (one wrapper for console tools, replacing four copies), Invoke-Utf8Command and Invoke-Git (UTF-8 output, Invoke-Git throws on a non-zero exit code), ConvertFrom-GitNameStatus (parses git diff -z --name-status), Get-VipmPath, Get-GhPath.
  • ConvertTo-MarkdownText (escapes & < > |, breaks @ mentions with a zero-width space) and Format-MarkdownCodeBlock (fence longer than any backtick run in the text).
  • Get-AnsiEncoding replaces [Text.Encoding]::Default for LabVIEW.ini and VIPM's Settings.ini (GetEncoding(0), 1252 as fallback); Add-GitHubFileText writes GITHUB_ENV/PATH/summary as UTF-8 without BOM.
• tools/ci/Resolve-CiEnvironment.ps1 (modified): the clean start runs after the test and comparison LabVIEWs are chosen and passes only those two. Closing the test LabVIEW follows the same rule, and the VIObjCache is kept while that LabVIEW still runs. Uses the shared lookups and UTF-8 writes.
• tools/ci/New-ViDiffReport.ps1 (modified): git worktree prune first; every git call checked; file list from git -c core.quotePath=false diff -z; LabVIEW's difference texts, frame captions and folder names escaped (tags dropped); CLI output in a safe fence; -LiteralPath.
• tools/ci/Publish-PrReport.ps1 (modified): every git call checked; the report commit is built with worktree add --no-checkout, write-tree and commit-tree, so no temporary branch exists; it is pushed by its id. When the checkout has no credentials, the push sends GH_TOKEN as the AUTHORIZATION header actions/checkout would store, through GIT_CONFIG_COUNT/KEY_0/VALUE_0 of that one git call (not on a command line, not in a config file). When a header is already configured, or for a local run without GH_TOKEN, git pushes as before. gh exit codes are checked on every call; comments are read with gh api --paginate (REST: page by page); only a marked comment of the current user (github-actions[bot] in Actions, whose token cannot read /user) is edited.
• tools/ci/Convert-CarayaReport.ps1 (modified): parses the report by its structure (callers, then tests at one depth with their asserts one level deeper; assert level = the depth most lines share, deeper lines belong to the current test; only test lines are split into label and VI), escapes cells, safe fence for the protocol output, reads UTF-8 with -LiteralPath, accepts not-run for both statuses.
• tools/ci/Invoke-CliTest.ps1, tools/ci/Test-InstalledPackage.ps1, tools/setup/Install-Dependencies.ps1 (modified): shared wrapper and lookups, -LiteralPath. Test-InstalledPackage.ps1 writes its initialize request to a temp file of its own and deletes it, and finds vipm under ProgramW6432. Install-Dependencies.ps1 keeps a VIPM_DESKTOP_LIVELINESS_TIMEOUT that is already set.
• tools/ci/release.workflow.yml (modified template): see the exact differences below.
• tools/ci/pull-request.workflow.yml (modified template): see below.
• tools/ci/README.md, tools/setup/README.md (modified): port resolution as Resolve-ViServerPort does it (VIPM's port first); git restore -- tests after a mass compile, with "check git status first", instead of git checkout -- .; the release template described as it is; the isolation claim corrected (the PR's VIs and tester run on the same runner; its own scripts are used while main lacks them); CI_DEDICATED_MACHINE documented.

No VI changed, so there is no code summary and no VI diff report.

Differences between the templates and the live workflows

tools/ci/release.workflow.yml against .github/workflows/release.yml:

1. Top-level permissions: contents: read (was contents: write).
2. build-and-release: environment: release and permissions: contents: write.
3. actions/checkout@v4 → @11d5960a326750d5838078e36cf38b85af677262 # v4.4.0 (both jobs), actions/upload-artifact@v4 → @ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2, softprops/action-gh-release@v2 → @3bb12739c298aeb8a4eeaf626c5b8d85266b0e65 # v2.6.2. These are the commits v4, v4 and v2 point to today (looked up with gh api repos/<owner>/<repo>/git/ref/tags/<tag>; all three are lightweight tags on commits), so the action code is unchanged.
4. build-and-release checkout: fetch-depth: 0.
5. New step "Check that the Tag Is on main" (only for v* tags): fetches main, fails unless git merge-base --is-ancestor HEAD origin/main.
6. Build step: TAG_NAME: ${{ github.ref_name }} in env:, the script reads $env:TAG_NAME (was "${{ github.ref_name }}"); [Console]::OutputEncoding set to UTF-8 before the git calls; a failed tag fetch stops the step; tag text from %(contents:subject)%0a%0a%(contents:body), trimmed (was %(contents)); version pattern '^v(\d+)\.(\d+)\.(\d+)(?:\.(\d+))?$' (was unanchored "v?(\d+)...").
7. Release notes: every $ doubled before -replace (this was already in the template on main, not in the live file).
8. VIP_PATH written with [IO.File]::AppendAllText as UTF-8 (was echo ... >> $env:GITHUB_ENV).
9. Comments added or reworded at the changed places; lines that contained only spaces are now empty.

tools/ci/pull-request.workflow.yml against .github/workflows/pull-request.yml:

1. Header comment on env: use and UTF-8 writes.
2. All four checkouts pinned to 11d5960… # v4.4.0 with persist-credentials: false; upload-artifact (both) pinned to ea165f8… # v4.6.2.
3. Step ids setup, install, recheck on the three setup steps.
4. "Check that Every Test VI Is Called" step (Test-TestsWired.ps1), and steps.wired.outcome in the last step. Already in the template on main, not in the live file.
5. status= lines to GITHUB_OUTPUT with [IO.File]::AppendAllText (was >>, UTF-16 in Windows PowerShell).
6. Run Protocol Tests: if: ${{ !cancelled() && steps.recheck.outcome == 'success' && steps.tests.outputs.status != '2' }} (was always() && ...).
7. Build Test Report: if: ${{ !cancelled() }}; outcomes and outputs through env:; a skipped step is passed as not-run; summary appended as UTF-8.
8. Build VI Diff Report: if: ${{ !cancelled() && steps.setup.outcome == 'success' && steps.recheck.outcome != 'failure' }}; base branch through env: BASE_REF.
9. Upload Reports and Post Report: if: ${{ !cancelled() }}. Post Report gets PR_NUMBER, HEAD_SHA, DIFF_OUTCOME through env: and builds the run URL from GITHUB_SERVER_URL/GITHUB_REPOSITORY/GITHUB_RUN_ID; when the diff was skipped, the footer starts with "VI diff: not run".
10. Last step: if: ${{ !cancelled() && (setup/install/recheck not success || tests/protocol failure || diff not success || wired failure) }}, all values through env:, an extra error line for failed setup steps and for a diff that did not run, "not run" instead of "skipped".
11. test-build: if: github.event.pull_request.head.repo.full_name == github.repository.
12. test-build build step: PR_NUMBER, HEAD_SHA through env:, run number from $env:GITHUB_RUN_NUMBER; summary and VIP_PATH written as UTF-8.
13. "Test Installed Package" step (Test-InstalledPackage.ps1), already in the template on main, not in the live file; its summary is now written as UTF-8 inside a four-backtick fence.
14. Comments added or reworded at the changed places; newline at the end of the file.

Verified

All on this machine, Windows PowerShell 5.1.26100, without LabVIEW and without stopping any process:

• Every .ps1/.psm1 in the repository parsed with [System.Management.Automation.Language.Parser]::ParseFile: 0 errors. All changed scripts are ASCII.
• Every run: block of both templates extracted and parsed with the same parser: 14 + 9 blocks, 0 errors. No ${{ }} left inside any run: block, every uses: pinned to a 40-character SHA with a version comment, no tabs, final newline.
• Select-LabVIEWToStop with mocked processes: clean job LabVIEW stopped, unsaved one kept (reason names the window), LabVIEW outside the job kept also with CI_DEDICATED_MACHINE=1, path compare ignores case, CI_DEDICATED_MACHINE only for 1 (9 + 3 checks).
• Stop-LeftoverVipm with Get-Process/Stop-Process replaced inside the module (guarded so the test aborts if a mock is not active, process ids that cannot exist on Windows): stops exactly VIPM and the clean job LabVIEW, keeps the unsaved and the developer LabVIEW, notes the kept one.
• Resolve-ViServerPort -Fix with mocks: unsaved LabVIEW not stopped, LabVIEW.ini not written, running port used; with CI_DEDICATED_MACHINE=1 and for a clean LabVIEW stopped and the port written; without -Fix unchanged (10 checks).
• Get-ProcessWindowTitle: the EnumWindows type compiles in PowerShell 5.1 and reads a process's windows (read only).
• Invoke-NativeCommand: stdout and stderr as text, exit code, no stop with ErrorActionPreference = Stop, missing program gives -1 and a "not recognized" line.
• Get-AnsiEncoding gives the same code page (1252) as Encoding.Default did; LabVIEW.ini round trip keeps a non-ASCII line; Add-GitHubFileText writes UTF-8 without BOM.
• ConvertTo-MarkdownText, Format-MarkdownCodeBlock, ConvertFrom-GitNameStatus (also on a real scratch repository with a rename and a file named with ü, [1], ' and @), Invoke-Git throws on a failing call.
• Publish-PrReport.ps1 functions taken from its syntax tree, GET requests only, against PR Show every case frame of changed VIs in the detailed diff report #29 with pages of 2: gh --paginate and REST paging return the same comments; author lookup through gh and REST, fallback to github-actions[bot] only inside Actions; marker plus author filter; push header passed through GIT_CONFIG_* and restored, not doubled when the checkout already has one, not used without GH_TOKEN (git push replaced by a recorder).
• Publish-PrReport.ps1 run end to end offline: local bare repository as origin, gh replaced by a stub that only records its arguments. Pushed a commit without parent containing README.md, images/..., report.md to pr-assets/pr-7; no local branch or worktree left; checkout clean; gh called for user, --paginate and one POST.
• Convert-CarayaReport.ps1 on the most recent real report of this machine (64 tests): same tests and results as the version on main; differences are the full assert labels (main cut them at " - ") and the escaping. On crafted reports: one deeper assert line no longer hides tests, | < > @ escaped in labels and checks, not-run for both statuses, fence longer than backticks in the protocol output.
• Release step on a scratch tag object with a signature block: the new format gives subject and body without the signature (%(contents) included it), ü read correctly with the UTF-8 console; the anchored pattern matches v1.2.3 and v1.2.3.4 and not test-1.2.3 (the old one did).

Not verified

• No workflow ran: the templates are not active, and no CI script was run for real (Resolve-CiEnvironment.ps1, New-ViDiffReport.ps1, Invoke-CliTest.ps1, Test-InstalledPackage.ps1, Publish-PrReport.ps1 against GitHub). The push with GH_TOKEN as header has not been tried against GitHub.
• The YAML of the templates was not parsed with a YAML parser: none is installed here (no PyYAML, ruamel, js-yaml or ruby). Only the structural checks above.
• No LabVIEW was involved: whether LabVIEW 2020/2026 marks unsaved changes with * in every window title, and that the check sees a real LabVIEW's windows, was not observed.
• New-ViDiffReport.ps1 escaping was not run against a real LabVIEW comparison report.
• PowerShell 7 is not installed; GetEncoding(0) there is not tested.
• The release environment, the tag guard and !cancelled() behaviour on GitHub.

Known rough edges

• A test or comparison LabVIEW without unsaved changes is still force-closed, and VIPM Desktop is still ended at the clean start. Unsaved changes in a VI whose window is not open cannot be seen; a * in a title for another reason keeps that LabVIEW running (the safe side).
• Test-InstalledPackage.ps1 still closes the test LabVIEW at its end (LabVIEW CLI CloseLabVIEW, then Stop-Process) without the unsaved check; it is documented as "do not point it at a LabVIEW someone works in" and was not in scope.
• %(contents:subject) folds the first paragraph of a tag message into one line. A tag message whose first lines are not followed by a blank line gets them joined.
• The guard runs for v* tags only; test-* tags build only a test package. Because the environment is set on the whole job, test-* builds also wait for approval.
• The release template's test job and its checkouts still persist credentials (not part of the findings).
• git worktree prune also runs when the scripts are used locally; it only removes worktree entries whose folders no longer exist.
• Gating: a failed package install still lets the VI diff run (diagrams then show ? for missing subVIs), but not the tests and protocol tests.

Maintainer steps

1. Merge this pull request first: the templates rely on the new scripts on main (not-run, the push with GH_TOKEN). The current live workflows keep working with the new scripts.
2. Create the environment release (Settings > Environments) with yourself as required reviewer, and optionally a deployment rule for tags v*. Do this before step 3: a workflow that names a missing environment creates it without protection.
3. Copy both templates on main:

   git switch main; git pull
   Copy-Item tools\ci\pull-request.workflow.yml .github\workflows\pull-request.yml
   Copy-Item tools\ci\release.workflow.yml .github\workflows\release.yml
   git add .github\workflows\pull-request.yml .github\workflows\release.yml
   git commit -m "Activate the hardened CI workflows"; git push

4. Add a tag ruleset for v* (Settings > Rules > Rulesets, target tags, v*): restrict creation, update and deletion, with only yourself in the bypass list.
5. Optional: on a runner nobody works on, set CI_DEDICATED_MACHINE=1 in the runner's .env file and restart the runner. Leave it unset on the development machine.

🤖 Generated with Claude Code

[github-actions]

Tests

✅ All 64 tests passed.

Per-test results

Test	VI	Result	Checks
Test 1: Initialize	Test Initialize.vi	✅ pass	Assert True: pass
Test 2: Initialized	Test Initialized.vi	✅ pass	Assert True: pass
Test 3: tools/list	Test tools list.vi	✅ pass	Assert True: pass
Test 4: prompts/list	test prompts list.vi	✅ pass	Assert True: pass
Test 5: resources/list	test resources list.vi	✅ pass	Assert True: pass
Test 6: tools call	test tools call.vi	✅ pass	Assert True: pass
Test 7: empty JSON object is an invalid request	test empty msg.vi	✅ pass	Assert True: pass
Test 8: call broken tool	test broken tool.vi	✅ pass	Assert True: pass
Test 6: tools call	test tools call array.vi	✅ pass	Assert True: pass
Test 9: GET is rejected with 405	test http get rejected.vi	✅ pass	GET -> 405 with Allow: POST and error -32600: pass
Test 10: Origin validation	test http origin.vi	✅ pass	Origin http://evil.example -> 403 with error -32600: pass Origin http://localhost:6274 -> 200 with result: pass
Test 11: MCP-Protocol-Version validation	test http protocol version.vi	✅ pass	MCP-Protocol-Version 1999-01-01 -> 400 with error -32022 (Unsupported protocol version, the MCP 2026-07-28 code in both eras): pass MCP-Protocol-Version 2025-06-18 -> 200 with result: pass
Test 12: notification answered with 202	test notification accepted.vi	✅ pass	notifications/initialized -> 202 with empty body: pass
Test 13: request with id answered with 200	test request status.vi	✅ pass	tools/list with id 1 -> 200 with result and id 1: pass
Test 14: unknown method	test unknown method.vi	✅ pass	does/not/exist with id 5 -> 200 with error -32601 and id 5: pass
Test 15: unknown notification	test unknown notification.vi	✅ pass	notifications/cancelled -> 202 with empty body: pass
Test 16: ping	test ping.vi	✅ pass	ping with id 3 -> 200 with result {} and id 3: pass
Test 17: string request id	test string id.vi	✅ pass	tools/list with id "abc-1" -> 200 with result and id "abc-1": pass
Test 18: unknown prompt	test unknown prompt.vi	✅ pass	prompts/get no-such-prompt with id 7 -> JSON-RPC error -32602 with id 7, no status/source: pass
Test 19: unknown resource	test unknown resource.vi	✅ pass	resources/read of an unknown uri with id 8 -> JSON-RPC error -32002 with id 8, no status/source: pass
Test 20: resources/read returns UTF-8 text unchanged	test resource text utf8.vi	✅ pass	text of labview://fixture_utf8 decodes to the bytes of fixture_utf8.txt: pass
Test 21: resources/read returns a binary file as Base64	test resource blob.vi	✅ pass	blob of labview://fixture_binary is the Base64 of the bytes of fixture_binary.png: pass
Test 22: unknown tool	test unknown tool.vi	✅ pass	tools/call no_such_tool -> error -32602 with message Unknown tool: no_such_tool and id 22: pass
Test 23: tools/call with arguments that are not an object	test tool call without arguments.vi	✅ pass	tools/call fake_tool_add with arguments 5 -> error -32602 with message Invalid arguments... and id 23: pass
Test 24: broken tool is a tool execution error	test broken tool message.vi	✅ pass	tools/call fake_tool_broken_vi -> result with isError true and a text that says not executable: pass
Test 25: prompts/get without arguments	test prompt without arguments.vi	✅ pass	prompts/get fake_prompt_concatanate_strings without arguments -> result with messages: pass
Test 26: request without method	test request without method.vi	✅ pass	a message with id 26 and no method -> error -32600 with id 26: pass
Test 27: malformed JSON	test malformed json.vi	✅ pass	truncated JSON -> error -32700 with id null: pass
Test 28: array inputs say what their items are	test tools list array items.vi	✅ pass	fake_tool_arrays: pass
Test 29: protocol version negotiation	test protocol version negotiation.vi	✅ pass	initialize asking for 2024-11-05 -> protocolVersion 2024-11-05: pass initialize asking for 1999-01-01 -> protocolVersion 2025-11-25 (the newest legacy version): pass
Test 30: resources/templates/list	test resource templates list.vi	✅ pass	resources/templates/list -> result with resourceTemplates []: pass
Test 31: tools/call without arguments	test tool call default arguments.vi	✅ pass	tools/call fake_tool_add without arguments -> result with the sum 0, no error, id 31: pass
Test 32: a tool that returns an error sets isError	test tool error result.vi	✅ pass	fake_tool_return_error with status true -> result with isError true, code and source still in the content: pass fake_tool_return_error with status false -> result without isError: pass
Test 33: internal errors show no LabVIEW internals	test internal error message.vi	✅ pass	prompts/get fake_prompt_broken_vi -> error -32603 with a message without call chain, markup, tags or path: pass
Test 34: prompts/list says which arguments are required	test prompts list required.vi	✅ pass	fake_prompt_required_argument: pass
Test 35: prompts/get checks required arguments	test prompt required argument.vi	✅ pass	prompts/get fake_prompt_required_argument without topic -> error -32602 that names topic, id 36: pass prompts/get fake_prompt_required_argument with topic -> result whose text contains both argument values: pass
Test 36: JSON-RPC batch	test batch.vi	✅ pass	batch of ping (id 38), a notification and tools/list (id b) -> 200 with an array of two answers in that order: pass batch of two notifications -> 202 with empty body: pass empty batch -> error -32600 with id null: pass
Test 37: a JSON value that is not an object	test non object message.vi	✅ pass	body 5 -> error -32600 with id null: pass body "text" -> error -32600 with id null: pass body null -> error -32600 with id null: pass
Test 38: tools/call checks required arguments	test tool required arguments.vi	✅ pass	tools/call fake_tool_required_input without arguments -> result with isError true and the text Missing required argument(s): b, a (the order of inputSchema.required), no JSON-RPC error, id 40: pass tools/call fake_tool_required_input with arguments {} -> result with isError true naming b and a, no JSON-RPC error, id 41: pass tools/call fake_tool_required_input with only a -> result with isError true naming only b, no JSON-RPC error, id 42: pass tools/call fake_tool_required_input with a 2 and b 3 -> result 6, no error, id 43: pass
Test 39: initialize with an unknown protocol version	test protocol version fallback.vi	✅ pass	initialize asking for 1999-01-01 -> protocolVersion is the first entry of Supported Protocol Versions.vi: pass
Test 40: initialize reports the LabVIEW version	test labview version.vi	✅ pass	initialize -> serverInfo.additionalDetails.labviewVersion equals the Version Number of this LabVIEW: pass
Test 41: tools/list shows the MCP names of special labels	test tools list mcp names.vi	✅ pass	tools/list fake_tool_special_labels -> Length (mm) and Factor #1 & 50% are listed as length__mm_ and factor__1___50_ in properties and in required, id 44: pass tools/list fake_tool_special_labels -> every name in properties and required matches ^[a-zA-Z0-9_-]{1,64}$: pass
Test 42: tools/call reaches special labels by their MCP names	test tools call mcp names.vi	✅ pass	tools/call fake_tool_special_labels with length__mm_ 2.5 and factor__1___50_ 4 -> result {"result__mm_":10}, no error, id 45: pass
Test 43: Add Tool refuses inputs whose MCP names clash	test add tool name clash.vi	✅ pass	Add Tool fake tool name clash.vi with the inputs a b and a_b -> error 5010 naming both controls and the MCP name a_b: pass tools/list after the refused Add Tool -> fake_tool_name_clash is not listed, id 46: pass
Test 44: Add Prompt refuses arguments whose MCP names clash	test add prompt name clash.vi	✅ pass	Add Prompt fake prompt name clash.vi with the arguments a b and a_b -> error 5010 naming both controls and the MCP name a_b: pass prompts/list after the refused Add Prompt -> fake_prompt_name_clash is not listed, id 47: pass
Test 45: 3D arrays and arrays of paths, enums and clusters say what their items are	test tools list array shapes.vi	✅ pass	fake_tool_array_shapes: pass fake_tool_array_shapes: pass fake_tool_array_shapes: pass fake_tool_array_shapes: pass
Test 46: timestamps, clusters and types without a JSON type are described in tools/list	test tools list schema details.vi	✅ pass	fake_tool_array_shapes: pass fake_tool_array_shapes: pass fake_tool_schema_shapes: pass fake_tool_schema_shapes: pass fake_tool_schema_shapes: pass fake_tool_schema_shapes: pass fake_tool_schema_shapes: pass
Test 47: tools/call converts timestamps, clusters, enums and 2D and 3D arrays	test tools call schema shapes.vi	✅ pass	fake_tool_array_shapes with a 3D array [[[1,2],[3,4]],[[5,6],[7,8]]] -> no isError, the tool gets all 8 elements: pass fake_tool_schema_shapes: pass fake_tool_schema_shapes: pass fake_tool_schema_shapes: pass fake_tool_schema_shapes with timestamp 2026-10-17T12:34:56Z -> no isError, time out 2026-10-17T12:34:56.000Z: pass
Test 48: every array schema in tools/list says what its items are	test tools list every array has items.vi	✅ pass	tools/list of the test server: pass
Test 49: tools/list gives element descriptions, integer ranges, data entry limits, units and defaults	test tools list descriptions defaults ranges.vi	✅ pass	fake_tool_schema_shapes: pass fake_tool_schema_details: pass fake_tool_schema_details: pass fake_tool_schema_details: pass fake_tool_schema_details: pass fake_tool_schema_details: pass fake_tool_schema_details: pass fake_tool_schema_details: pass
Test 50: tools/call returns structuredContent without picture outputs, and none on isError results	test tools call structured content.vi	✅ pass	tools/call fake_tool_required_input without arguments (id 503) -> isError true and no structuredContent: pass tools/call fake_tool_return_error with status true (id 501) -> isError true and no structuredContent: pass tools/call fake_tool_return_error with status false (id 502) -> structuredContent carries the error cluster output as an ordinary object: pass fake_tool_picture_outputs: pass fake_tool_picture_outputs: pass tools/call fake_tool_picture_outputs (id 50) -> structuredContent is present and equals the text JSON without its picture placeholder: pass
Test 51: tools/list gives an outputSchema that matches structuredContent	test tools list output schema.vi	✅ pass	tools/list fake_tool_no_inputs_outputs (no outputs) -> no outputSchema: pass tools/list fake_tool_picture_outputs -> outputSchema is an object schema whose properties and required are count, label, mode and time: pass fake_tool_picture_outputs: pass fake_tool_picture_outputs: pass fake_tool_picture_outputs: pass
Test 52: tools/list sends title and annotations only when they are set	test tools list title annotations.vi	✅ pass	tools/list fake_tool_add (registered without title and annotations) -> neither title nor annotations: pass tools/list fake_tool_annotated (registered with a title) -> title "Annotated Fixture Tool": pass fake_tool_annotated (readOnlyHint true, destructiveHint false, the others not set) -> annotations with exactly readOnlyHint true and destructiveHint false: pass
Test 53: modern results carry resultType, cache hints and serverInfo; legacy results do not	test modern tools list.vi	✅ pass	modern tools/list -> _meta io.modelcontextprotocol/serverInfo with name and version: pass modern tools/list (id 53, MCP 2026-07-28 headers and _meta) -> 200 with tools, resultType complete, ttlMs 60000 and cacheScope public: pass legacy tools/list (id 531, no _meta) -> tools without resultType, ttlMs, cacheScope or _meta: pass
Test 54: server/discover, and initialize never negotiates 2026-07-28	test server discover.vi	✅ pass	initialize asking for 2026-07-28 (id 541) -> the newest legacy version 2025-11-25, never 2026-07-28: pass server/discover -> capabilities tools, prompts and resources (as initialize), _meta serverInfo, ttlMs 60000, cacheScope public: pass server/discover (id 54) -> 200, resultType complete, supportedVersions with 2026-07-28 and 2025-11-25: pass initialize asking for 2025-11-25 (id 542) -> 2025-11-25: pass
Test 55: modern requests are refused when their headers do not match the body	test modern headers.vi	✅ pass	tools/call with an Mcp-Name that is not params.name (id 554) -> 400 with error -32020: pass MCP-Protocol-Version 2025-11-25 while _meta says 2026-07-28 (id 553) -> 400 with error -32020: pass tools/call without an Mcp-Name header (id 551) -> 400 with error -32020: pass tools/call with Mcp-Name =?base64?...?= of fake_tool_add, lower-case header names (id 552) -> 200 with the tool result: pass Mcp-Method prompts/list on a tools/list request (id 55) -> 400 with error -32020: pass
Test 56: unsupported versions, missing capabilities and unknown methods in the modern era	test modern errors.vi	✅ pass	modern resources/read of an unknown uri (id 565) -> error -32602 (legacy keeps -32002): pass modern request with the unknown method foo/bar (id 563) -> 404 with error -32601: pass modern request with the legacy version 2025-11-25 in _meta (id 561) -> 400, error -32022, data.supported with 2025-11-25: pass unknown version 2027-01-01 in header and _meta (id 56) -> 400, error -32022, data.supported with 2026-07-28, data.requested 2027-01-01: pass modern ping (id 564), which MCP 2026-07-28 removed -> 404 with error -32601: pass modern request without io.modelcontextprotocol/clientCapabilities (id 562) -> 400 with error -32602: pass
Test 57: elicitation by input_required and retry with inputResponses	test elicitation.vi	✅ pass	legacy tools/call (id 574) -> outcome not supported: pass retry with inputResponses confirm_action accept (id 571) -> complete result with outcome confirmed: pass modern tools/call of fake_tool_confirm with the elicitation capability (id 57) -> resultType input_required with inputRequests.confirm_action: pass modern tools/call without the elicitation capability (id 573) -> outcome not supported, no input_required: pass retry with inputResponses confirm_action decline (id 572) -> outcome declined: pass
Test 58: progress notifications as Server-Sent Events for tools/call	test progress notifications.vi	✅ pass	legacy tools/call with progressToken "t58" and Accept text/event-stream (id 581) -> HTTP 200, text/event-stream: pass id 581: exactly 10 notifications/progress events, each with progressToken "t58" and total 10, progress 1..10 in increasing order: pass the tool finished all 10 steps (fixture record): pass id 581: the JSON-RPC result is the last data event, after progress 10: pass the same call without progressToken (id 582) -> plain application/json, no progress events: pass modern tools/call with the numeric progressToken 583 (id 583) -> text/event-stream, 10 events with "progressToken":583 in order, then the complete result: pass
Test 59: closing the stream cancels a streamed tools/call	test progress cancel by close.vi	✅ pass	modern streamed tools/call over raw TCP (id 591) -> text/event-stream, the first progress event arrives: pass after the client closed the connection the tool stopped early (fewer than 10 steps) and saw cancelled? TRUE: pass
Test 60: legacy notifications/cancelled stops a streamed tools/call	test progress legacy cancel.vi	✅ pass	legacy streamed tools/call over raw TCP (id 601) -> text/event-stream, the first progress event arrives: pass the tool stopped early (fewer than 10 steps) and saw cancelled? TRUE: pass the stream ended without a result for id 601: pass notifications/cancelled for id 601 on a second connection -> HTTP 202 within 1 s, while the call runs: pass
Test 61: a client that is not in Allowed clients gets 403	test access allowed clients.vi	✅ pass	tools/list from this computer to a server that allows only 192.0.2.1 -> HTTP 403, application/json: pass the second server (port 37061, Allowed clients = [192.0.2.1]) started: pass the 403 body is the JSON-RPC error -32600 "Forbidden: client not allowed." with id null: pass
Test 62: Access token requires Authorization: Bearer	test access bearer token.vi	✅ pass	the token never appears in a response: pass the second server (port 37062, Access token set) started: pass authorization: Bearer test-token-62 (lower-case header name) -> HTTP 200 with the tools/list result: pass the 401 body is the JSON-RPC error -32600 "Unauthorized: missing or invalid bearer token." with id null: pass Authorization: Bearer wrong-token -> HTTP 401 with WWW-Authenticate: Bearer: pass no Authorization header -> HTTP 401 with WWW-Authenticate: Bearer (header names compared case-insensitively): pass Authorization: bearer test-token-62 (lower-case scheme) -> HTTP 200 with the tools/list result, no WWW-Authenticate: pass
Test 63: tools/list leaves out LabVIEW's default error cluster descriptions	test tools list error descriptions.vi	✅ pass	fake_tool_error_descriptions: pass tools/list fake_tool_error_descriptions -> error in (no error) and error out are in inputSchema and outputSchema without a description key, and so are their status, code and source; their type is still there: pass

Protocol tests

✅ Test-McpServer.ps1 passed against the test server.

Protocol tester output (0 failed required checks)

Testing MCP server at http://127.0.0.1:36987/mcp/server

[PASS] (required) Lifecycle - initialize answers 200 with a JSON result
[PASS] (required) Lifecycle - initialize answer is JSON-RPC 2.0 with the request id
[PASS] (required) Lifecycle - initialize result has protocolVersion, capabilities and serverInfo.name
[PASS] (required) Lifecycle - Content-Type of the answer is application/json
[PASS] (advisory) Lifecycle - a client asking for 2024-11-05 gets that version back
[PASS] (required) Lifecycle - notifications/initialized is accepted with 202 and no body
[PASS] (required) Transport - GET answers 405 or opens an event stream
[PASS] (advisory) Transport - 405 answer names the allowed method in an Allow header
[PASS] (required) Transport - a request from a foreign Origin is rejected (403)
[PASS] (required) Transport - a request from a localhost Origin is accepted
[PASS] (required) Transport - an unsupported MCP-Protocol-Version header is rejected (400)
[PASS] (required) Transport - the negotiated MCP-Protocol-Version header is accepted
[PASS] (required) JSON-RPC - ping is answered with an empty result
[PASS] (required) JSON-RPC - an unknown method gets error -32601 with the request id
[PASS] (required) JSON-RPC - an unknown notification is not answered
[PASS] (required) JSON-RPC - a string request id is returned unchanged
[PASS] (required) JSON-RPC - malformed JSON gets error -32700
[PASS] (advisory) JSON-RPC - a request without a method gets error -32600
[PASS] (required) Tools - tools/list returns a tools array
[PASS] (required) Tools - every tool has a name and an inputSchema of type object (1 tools)
[PASS] (advisory) Tools - array parameters say what their items are
[PASS] (required) Tools - calling an unknown tool is answered with an error
[PASS] (required) Prompts - prompts/list returns a prompts array
[PASS] (required) Prompts - every prompt has a name (1 prompts)
[PASS] (required) Prompts - getting an unknown prompt is answered with an error
[PASS] (required) Resources - resources/list returns a resources array
[PASS] (required) Resources - every resource has a uri and a name (2 resources)
[PASS] (required) Resources - reading an unknown resource is answered with an error
[PASS] (advisory) Resources - resources/templates/list returns a resourceTemplates array

Required: 24 of 24 passed. Advisory: 0 failed.

VI diff report

No VI, VIM or CTL files changed in this pull request.

Report for commit f354e18. The installable test package is under Artifacts of this run once the build job has finished.