CHANGELOG.md

Changelog

All notable changes to modelchoice-mcp. Versions are tag-driven; pushing a vX.Y.Z tag publishes to PyPI via the release workflow.

Unreleased

• The wheel is now installed and booted before it is published (AB#3143). The wheel is what PyPI serves and what pip install gives you, but nothing ever installed it — the exe check gates the PyInstaller binary built from source, a different artifact with different dependency resolution, so a wheel-only fault (a missing packages entry, a bad hatch build glob, an undeclared dependency) would have shipped. The build job now installs the freshly-built wheel into a clean virtualenv, makes it answer a real MCP initialize handshake, and asserts the version it reports matches the tag being released — so a forgotten version bump can no longer burn a PyPI version number.
• The PyPI publish now waits for the Windows exe verification (AB#3142). publish-pypi depended on build alone, so build-windows-exe — which packs the single-file exe and asserts it answers an MCP initialize handshake — raced the publish instead of gating it, and a packaging break would still have reached PyPI. Publishing is the one irreversible step in the workflow (a version number can never be reused), so it now runs after every verification. Costs a few minutes per release; buys the ability to fix a bad build by deleting a tag rather than burning a version. Release-workflow only — no change to the package.

0.0.31

• Migrated to the mcp 2.0 SDK (AB#3134). mcp 2.0 removed mcp.server.fastmcp and renamed the high-level server class FastMCP → MCPServer (from mcp.server import MCPServer). The @mcp.tool / @mcp.resource / @mcp.prompt decorators are unchanged and still return the plain function, so all 25 tools, 4 resources and 2 prompts register exactly as before — verified name-for-name against the 1.x baseline. The dependency is now mcp>=2,<3; the major-version cap stays, deliberately.
  • HTTP transports: host and port are now run() keyword arguments (mcp.run(transport=..., host=..., port=...)) instead of mutating mcp.settings beforehand. --transport, --host and --port behave as before.
  • The server now reports its own version in serverInfo — MCPServer takes a version argument, which FastMCP had no equivalent for.

0.0.30

• CI stops drifting with upstream releases. uv.lock is now committed (it was in .gitignore), and CI syncs with --locked. Previously CI had no pins at all: it re-resolved the open floors in pyproject.toml on every run, so a new upstream major landed the day it shipped — mcp 2.0.0 moved mcp.server.fastmcp and broke mypy across every @mcp.tool on an unchanged main. mcp is also capped to <2 for people installing from PyPI, who resolve against the constraints rather than the lock. Editing pyproject.toml now requires re-running uv lock and committing the result; CI fails with a clear message if you forget, instead of silently upgrading you.
• export_tree_json now carries a generator field (AB#3123) — product (ModelChoice by Vose Software), server version, UTC export timestamp and the product URL, so an exported tree stays attributable once it lands in version control, a ticket, or someone else's repository. It sits alongside model_json, never inside it: model_json still round-trips byte-identically through import_tree_json. Part of the ModelChoice output-branding work (Feature AB#3118).

0.0.29

• Licence gate (AB#2659) — building and analysis ACTIONS now require a fully licensed ModelChoice. The bridge reads the add-in's licence state via the new headless MC_LicenseStatus_Auto and refuses actions (build/edit commit, build_mcda, control panel, set_input_distribution, run_utility / evii / evpi / risk_profile / decision_report / robustness / sensitivity / analysis, import) unless isComplete (full licence). Reading is unaffected — list/get/roll_up/ verify/scenarios/export and open/close workbook work regardless. New read-only license_status tool reports the state. Fail-closed: if the status can't be read (add-in missing/old), actions are blocked with an actionable message. (Trial/expired users can read but not drive actions.) Needs the add-in build with MC_LicenseStatus_Auto.
• .mcpb now built with the official mcpb CLI (@anthropic-ai/mcpb, validates during pack) instead of a hand-rolled zip; plain-zip fallback when node isn't present. README install section reworked: PyPI + config is the recommended path; the .mcpb one-click carries a note that the Claude Desktop Extensions installer silently no-ops on the latest Windows MSIX builds (a client bug, not the bundle) — use pip install until Anthropic patches it.

0.0.28

• One-click install: Claude Desktop Extension (.mcpb) — the release now also builds a standalone Windows .exe (new PyInstaller spec) and wraps it in a .mcpb Desktop Extension attached to the GitHub release. Installing into Claude Desktop becomes 1. open modelchoice-mcp.mcpb 2. restart Claude — no Python, no claude_desktop_config.json editing. Built CLI-free in CI (scripts/build_mcpb.py); the bundle version is injected from the release tag. Excel + the ModelChoice add-in are still required for rendering.

0.0.27

• close_workbook — close an open workbook by file name (counterpart to open_workbook). By default unsaved changes are discarded (save=False); pass save=True to write them first. (Mirrors close_workbook in modelrisk-mcp.)
• open_workbook suppresses Excel's open prompts — Workbooks.Open is now called with update_links=False, ignore_read_only_recommended=True, notify=False, add_to_mru=False and with DisplayAlerts/AskToUpdateLinks off, so a headless open can't hang on an Update-Links / read-only dialog. (External links aren't refreshed on open; values stay as last saved.)

0.0.26

• open_workbook — open a decision-tree workbook (.xlsx) from disk in the running Excel so the other tools can act on it. Reports the workbook's sheets and any ModelChoice tree sheets it contains; reuses an already-open workbook of the same name. (Mirrors the new open_workbook in modelrisk-mcp.)

0.0.25

• build_mcda gains AHP weight elicitation (AB#2646) — set weight_source="ahp" and pass an ahp_matrix (Saaty 1-9 pairwise comparisons, (1 + criteria) square, ordered [financial, …criteria]) instead of direct weights. The tool computes the eigenvector weights + consistency ratio for the preview and forwards weightSource/ahpMatrix to MC_ApplyMcda_Auto, which recomputes them authoritatively (CR surfaced; CR > 0.10 flagged inconsistent). McdaBuildResult now reports weight_source, weights, consistency_ratio.

0.0.24

• /decision-tree-monte-carlo prompt — guides the cross-server Monte Carlo hand-off: assign Vose* distributions to a tree's uncertain inputs (set_input_distribution), wrap the root EV (MC_V_<rootId>) as a ModelRisk output, run the simulation in modelrisk-mcp, and read the EV's output distribution back. The /design-decision-tree prompt's step 5 now points to it.
• README roadmap refreshed — Phases 0–3 marked delivered (22 tools, 2 prompts), with simulation-orchestration and AHP-for-MCDA as candidate next steps.

0.0.23

• set_input_distribution — assign an uncertainty (a ModelRisk Vose* distribution) to a tree input (branch cash flow or probability), the way the UI lets you type a distribution into a cell. Stored as the input's user-formula and re-rendered, so it persists. This is the decision-tree half of a Monte Carlo: once inputs are distributions, run the simulation with modelrisk-mcp (it samples these cells and collects the tree's output distribution). Pure _MC_Store edit + re-render — no new add-in command.

0.0.22

• build_mcda — build a multi-criteria (MCDA) model: tree + criteria (ordinal options, weights, direction) + aggregation + per-terminal scores. Validates structure/weights/scores in Python; drives the new MC_ApplyMcda_Auto (AB#2637) to set MCDA mode + render. v1 = direct weights (AHP follow-up). No-op until that add-in build ships.

0.0.21

• run_utility — risk-attitude (utility) rollback: certainty equivalent, risk premium, and the optimal decision under risk aversion vs EV. Drives the new MC_Utility_Auto (AB#2635); no-op until that add-in build ships.

0.0.20

• Drivers for the C# add-in batch — import_precisiontree (drives MC_ImportPrecisionTree_Auto) and two_way_sensitivity via run_decision_report (drives MC_TwoWaySensitivity_Auto). Both no-op until the matching add-in build (PR !2032) is installed. (Terminal-payoff linking in the control panel is add-in-side; no new MCP tool needed.)

0.0.19

• Housekeeping — CHANGELOG; gated live integration tests (MODELCHOICE_LIVE=1) exercising the real COM attach/read/rollback path.

0.0.18

• Knowledge resources — modelchoice://guide/* curated decision-analysis guidance (building trees, EVPI vs EVII, which analysis to use, pitfalls).

0.0.17

• Import / export — export_tree_json (a tree's raw ModelChoice JSON, to save/share/version) and import_tree_json (validate raw JSON → write/render).

0.0.16

• run_decision_report gains force_to_outcome (what inputs would have to change to force a chosen outcome).

0.0.15

• run_risk_profile — per-option outcome distribution (expected value, min, max, std dev + the cumulative-probability table), not just the EV.

0.0.14

• run_scenarios — what-if comparison: named bundles of input edits, each rolled back and compared to the baseline (EV, optimal decision, Δ, flips).

0.0.13

• Excel attach fix — bind to the running Excel via GetActiveObject (ROT) instead of xlwings' window-handle walk, which failed (0x800A01A8) once the ModelChoice add-in was loaded. Fixes timeouts/hangs from the MCP server.

0.0.12

• build_control_panel — lift a tree's inputs into a labelled control panel at the top of its sheet, with the tree linked back to it.

0.0.11

• edit_tree gains structural ops: add_option, add_branch, remove_branch.

0.0.10

• run_evii — Expected Value of Imperfect Information for a specific test.

0.0.9

• run_decision_report — strategy table / policy / decision brief / MCDA.

0.0.1 – 0.0.8

• Phases 0–2: read + rollback engine (list_trees, get_tree, roll_up, verify_rollback), build/edit (build_tree, edit_tree), and analysis drivers (run_evpi, run_robustness, run_sensitivity, run_analysis, read_sheet), plus the /design-decision-tree prompt. CI + tag-driven PyPI release. First PyPI publish at 0.0.1.