v0.3.0-alpha.8: e2e integration test passes live; document launch-order caveat

The 7 integration tests in tests/integration/test_e2e_run_simulation.py
all pass against a real Excel + ModelRisk XLL + MRService.dll round-trip
on a developer machine (18.85s, including the actual Monte Carlo run).
Empirical moments of Y = 2*N(0,1) match the analytic ones inside the
documented tolerance bands. First validation that the v0.3 pipeline
works end-to-end against the real stack.

The test run surfaced one real operational caveat — now documented in
the README's "Known caveats" section:

  Excel must be running interactively (Start menu / taskbar) before
  the MCP server tries to drive run_simulation. When Excel is launched
  programmatically by an automation client, ModelRisk's XLL skips part
  of its xlAutoOpen initialisation. XLL functions still register as
  worksheet UDFs (so VoseNormal(0,1) in a cell works), but XLL commands
  (VoseStartSimulCustom12 etc.) never get added to Excel's
  Application.Run table — and run_simulation depends on those commands.
  If run_simulation fails with "macro may not be available", restart
  Excel by hand and try again.

This is a ModelRisk XLL behaviour, not a server bug, but the launch-
order requirement is operationally important. The earlier ad-hoc test
worked because Excel had been open interactively.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: vosesoft

CHANGELOG.md

@@ -4,6 +4,18 @@ All notable changes to ModelRisk MCP. Follows [Keep a Changelog](https://keepach
 
 ## [Unreleased]
 
+## [0.3.0-alpha.8] — 2026-05-21
+
+End-to-end integration test passed live for the first time, surfacing one operational caveat now documented.
+
+### Verified
+
+- All 7 integration tests in `tests/integration/test_e2e_run_simulation.py` pass against a real Excel + ModelRisk XLL + MRService.dll round-trip (18.85s total). The empirical moments of `Y = 2 * N(0, 1)` over 1000 iterations match the analytic moments inside the documented tolerance bands. All v0.3 read-path tools (`list_vmrs_variables`, `get_samples`, `diagnose_workbook`) work end-to-end.
+
+### Documented
+
+- **Launch order caveat in the README's "Known caveats" section.** Excel must be running interactively (Start menu / taskbar) before the MCP server tries to drive `run_simulation`. When Excel is launched programmatically by an automation client, ModelRisk's XLL skips part of its `xlAutoOpen` initialisation — XLL functions still register as worksheet UDFs (so cell formulas work), but XLL commands (`VoseStartSimulCustom12` etc.) never get added to Excel's `Application.Run` table, and the simulation pipeline depends on those commands. This is a ModelRisk XLL behaviour, not a bug in this server, but the launch-order requirement is now explicit. The integration test discovered this; the earlier real-workbook test by hand worked because Excel had been open interactively.
+
 ## [0.3.0-alpha.7] — 2026-05-21
 
 Hotfix for the MCP Registry publish step that failed in 0.3.0-alpha.6.

README.md

@@ -217,6 +217,7 @@ More: [docs/architecture.md](docs/architecture.md), [docs/com-surface.md](docs/c
 
 ## Known caveats
 
+- **Launch order: Excel first, MCP server second.** Excel must already be running interactively (started from the Start menu, taskbar, or by double-clicking an `.xlsx`) before the MCP server tries to drive `run_simulation`. When Excel is launched programmatically by an automation client, ModelRisk's XLL skips part of its `xlAutoOpen` initialisation — the XLL functions still register as worksheet UDFs, but the XLL commands (`VoseStartSimulCustom12` etc.) never get added to Excel's `Application.Run` table. The simulation pipeline depends on those commands. If `run_simulation` fails with "macro may not be available", restart Excel by hand and try again.
 - **OneDrive-hosted workbooks**: xlwings can fail to resolve the workbook's full path without `ONEDRIVE_COMMERCIAL_WIN` set. The bridge degrades gracefully — name-based operations still work, and `run_simulation` defaults the `.vmrs` save location to the user's Desktop when the workbook folder can't be resolved.
 - **Active simulation results**: `get_simulation_results` reads from the `.vmrs` file produced by the most recent `run_simulation` call, or the most recent sibling `.vmrs` next to the workbook. Use `set_active_vmrs(path)` or `read_vmrs(path)` to point at a specific file.
 

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "modelrisk-mcp"
-version = "0.3.0a7"
+version = "0.3.0a8"
 description = "Open MCP server bridging Anthropic Claude (and any MCP-compatible client) with the ModelRisk Excel add-in."
 readme = "README.md"
 requires-python = ">=3.11"

server.json

@@ -3,12 +3,12 @@
   "name": "io.github.vosesoftware/modelrisk-mcp",
   "title": "ModelRisk",
   "description": "Read, build, fit, and run Monte Carlo risk models in Excel through Vose Software's ModelRisk.",
-  "version": "0.3.0a7",
+  "version": "0.3.0a8",
   "packages": [
     {
       "registryType": "pypi",
       "identifier": "modelrisk-mcp",
-      "version": "0.3.0a7",
+      "version": "0.3.0a8",
       "transport": {
         "type": "stdio"
       }

src/modelrisk_mcp/__init__.py

@@ -1 +1 @@
-__version__ = "0.3.0a7"
+__version__ = "0.3.0a8"

tests/unit/test_server_boot.py

@@ -5,7 +5,7 @@
 
 
 def test_version_is_set() -> None:
-    assert __version__ == "0.3.0a7"
+    assert __version__ == "0.3.0a8"
 
 
 def test_server_name() -> None: