davinci-resolve-mcp 2.88.0 → 2.90.0

package/CHANGELOG.md

@@ -2,6 +2,149 @@
 
 Release history for the DaVinci Resolve MCP Server. The latest release is summarized in the root README; older entries live here to keep the README focused.
 
+## What's New in v2.90.0
+
+AAF turnovers parsed by `editorial.parse_interchange` on the advanced server now
+carry their animation curves, the transforms nobody had interpreted, and the
+effects that occupy record time while emitting nothing. All three were losses a
+consumer could not see, because the parse reported itself complete.
+
+### Added
+
+- **Keyframe curves ship instead of the word "varying".** A `VaryingValue` was
+  read for its values alone, and only to answer "one number or more than one" —
+  `ControlPoint.time` was never asked for, so every animated reframe reached a
+  consumer as `"varying"`: enough to refuse the clip, never enough to rebuild
+  it. Transform stages now carry `keyframes[<AvidParamName>]` and retimes carry
+  `speedCurve.playRate` / `speedCurve.sourceOffset`, each with its interpolation
+  and its points as `{t, v, frame}`.
+- **`domain` names which rule produced `frame`,** because the two curve families
+  do not share a time domain and a single conversion rule would have been wrong
+  by an effect's whole length on one of them. Measured over all 2047 control
+  points of an 878-event Avid picture turnover: transform params are normalized
+  over the effect span with the endpoint inclusive (`frame = t x (length - 1)`,
+  which lands 1243 of 1254 points on an integer frame against 278 under
+  `length`), while speed maps are already in frames. Keys outside `0..1` are
+  kept rather than clamped — 107 of 1254 sit before the first frame or past the
+  last, which is what Avid leaves when an animated clip is trimmed. A curve with
+  one unreadable point is refused whole; interpolating through a missing key
+  produces a confident wrong animation.
+- **`passthrough` stages carry the four uninterpreted transform operations**
+  (SBlend_v2, Stabilize_2, MaskImage_2, 2DMatteKey_2) that were previously
+  discarded whole — 22 events on the fixture. Their numbers travel in
+  `rawParams`, deliberately not `params`: the same parameter *name* carries
+  different units on different operations (SBlend's `DVE_POS_X_U` runs to -315
+  where Stabilize's runs to -0.92 on the same show), so there is nothing to
+  normalize, and a 2DMatteKey `AFX_POS_X_U` of 500 promoted into `params` would
+  become a ~960px shift of a clip nobody repositioned.
+- **`effectsWithoutEvents` closes a hole `unhandled` structurally cannot see.**
+  An effect can be modelled perfectly and still emit nothing — the group is
+  walked, its inputs are walked, and they contain no `SourceClip`. On the
+  fixture `unhandled` reads `{}` (a complete parse) while 29 SubCap titles
+  occupy real record time and reach the consumer as nothing whatsoever. Charged
+  once, to the innermost cause.
+- **`speedRatioFromCurve`** — the rate the offset curve itself implies, emitted
+  only when that curve is straight, so no variable timewarp is ever averaged
+  into a single number.
+
+### Fixed
+
+- **The declared AAF `SpeedRatio` rational is the source span truncated to whole
+  frames, and 7 of 18 constant retimes on the fixture disagree with their own
+  curve** — `201/112 = 1.794643` where the curve says `201.6/112 = 1.80`, and at
+  worst `31/19 = 1.631579` against a curve reading `1.70`, a 4% speed error in a
+  number an operator is handed to type in by hand. Every curve slope lands on a
+  rate an editor would actually dial (1.7, 1.8, 2.0, 0.75); every declared value
+  is that rate spoiled by rounding. Both ship under their own names and neither
+  is substituted for the other.
+- **Variable timewarps are reconstructible.** The offset curve is dense (up to
+  387 points at half-frame steps) and describes where every record frame reads
+  from, taking the 4 variable timewarps on the fixture from "flagged, rebuild
+  from nothing" to fully described — including one reverse ramp whose offset
+  runs 301.0 to -0.78. Reverse read off the curve agreed with the declared flag
+  9 of 9, in both directions.
+
+### Validation
+
+- Every pre-existing field is byte-identical on the fixture, verified by a
+  structural diff of the full 878-event parse against the previous output; event
+  count, `unhandled`, and all existing counters are unchanged.
+- 10 new tests in `resolve-advanced/test/aaf-sequences.test.mjs`.
+- No DaVinci Resolve scripting behavior changed; this is offline interchange
+  parsing, so no live Resolve validation was required.
+
+## What's New in v2.89.0
+
+The build gates this server already enforced are now gates an agent can ask
+about, and the capability probe those gates run on no longer lies. Issue #132,
+reported by @magwa101.
+
+### Added
+
+- **`get_version` reports what the connected build is missing.** A new `build`
+  block carries `unavailable_on_this_build` — every recorded API surface this
+  build does not have — plus `known_gates` and the caveat that an absence from
+  that list is not a promise a method exists. #132 happened in a session that
+  opened with `get_version` and was told a number and nothing the number ruled
+  out. `get_resolve_version_fields` gains the same list on the granular server.
+- **The version-gate registry went from 7 recorded surfaces to 41.**
+  `_requires_method(obj, "GetLayoutPresetList", "21.0.4")` appears 44 times
+  across the two servers and gates 31 distinct symbols, but
+  `check_version_support` — the call `resolve-session` step 2 tells an agent to
+  make — answered from a ledger that knew seven of them. So the one question an
+  agent is instructed to ask returned `unknown` for surfaces this server was
+  already routing on. On Studio 19.1.3.7 the session preflight named 7 missing
+  surfaces; it now names 40. The new floors are labelled `documented` rather
+  than `measured`: they come from Blackmagic's release documentation, not a live
+  bisect here, and an agent relaying one should be able to say which.
+- **`tests/test_version_gate_drift.py`** fails when a call site and the ledger
+  disagree, and when a bare method name is gated at two different builds on two
+  different classes (which would make a bare-name lookup a coin flip). A table
+  copied by hand is exactly what drifts back apart.
+- **`src/utils/resolve_probe.py`** — `has_method` and `api_constant`, with the
+  measurement behind them. A companion guard fails the suite on a bare `hasattr`
+  with a Resolve-shaped attribute name.
+
+### Fixed
+
+- **29 capability probes across `src/` used `hasattr` on Resolve API objects,
+  where it is a constant `True`** (measured on Studio 19.1.3.7 across 42 checks,
+  recorded in `api_truth`; re-confirmed live for this release —
+  `hasattr(project, 'GenerateSpeech')` returns `True` on a build that has no
+  such method). It failed in two shapes that look nothing alike:
+  - `if not hasattr(clip, "RemoveMotionBlur")` is a dead branch, so the
+    "requires Resolve 21+" refusal never fired and the call below it raised
+    `AttributeError` on an older build. Eleven of these were the granular
+    server's Resolve 21 AI guards.
+  - `getattr(r, n) if hasattr(r, n) else n` never reaches its `else`, so a build
+    without the constant got `None` where the author wrote a string fallback —
+    no exception, no refusal, just a `None` travelling on into `Export()`. This
+    reached granular timeline export and `ExportLUT`, and three constant
+    lookups on the compound server. Confirmed live: a name Resolve does not
+    define returned `None` under the old form and now falls back correctly.
+    The fallback keys on `is None`, not truthiness, because `EXPORT_AAF` is
+    genuinely `0.0`.
+- **The granular AI tools reported a missing Extras pack as success.** Resolve's
+  AI methods return the reason as a *string* when the pack is absent, and
+  `bool("Required package ... is not installed.")` is `True`. The compound
+  server has normalized that since the 21.0.2.4 measurement; the granular one
+  had not, so folder and clip audio classification returned `{"success": true}`
+  for a call that did not run, and `RemoveMotionBlur` / `GenerateSpeech` walked
+  a string into `.GetName()`. All now route through `_ai_result`, which reports
+  the failure and carries Resolve's own reason.
+- **`Project.ApplyFairlightPresetToCurrentTimeline` was recorded on `Timeline`.**
+  The shipped README lists it under Project and the server calls it there; the
+  method's name is what made the wrong attribution look right.
+
+### Validation
+
+Suite 2560 → 2580. Live-checked against the running Studio 19.1.3.7 for the
+probe semantics, the refusal path, the constant fallback, and the `get_version`
+preflight (40 of 41 gates unavailable, as expected on that build). The Resolve 21
+and 21.0.4 surfaces themselves remain untested here — this machine cannot run
+them — and the Extras-pack failure paths are pinned by a stub that reproduces
+the measured attribute-fabrication behaviour rather than by a live 21 build.
+
 ## What's New in v2.88.0
 
 The twelve Resolve 21.0.4 surfaces that only existed on the compound server now

package/README.md

@@ -2,7 +2,7 @@
 
 English | [简体中文](README.zh-CN.md)
 
-[![Version](https://img.shields.io/badge/version-2.88.0-blue.svg)](https://github.com/samuelgursky/davinci-resolve-mcp/releases)
+[![Version](https://img.shields.io/badge/version-2.90.0-blue.svg)](https://github.com/samuelgursky/davinci-resolve-mcp/releases)
 [![npm](https://img.shields.io/npm/v/davinci-resolve-mcp.svg?label=npm&color=CB3837)](https://www.npmjs.com/package/davinci-resolve-mcp)
 [![API Coverage](https://img.shields.io/badge/API%20Coverage-100%25-brightgreen.svg)](docs/reference/api-coverage.md)
 [![Tools](https://img.shields.io/badge/MCP%20Tools-34%20(353%20full)-blue.svg)](#server-modes)

package/README.zh-CN.md

@@ -2,7 +2,7 @@
 
 [English](README.md) | 简体中文
 
-[![Version](https://img.shields.io/badge/version-2.88.0-blue.svg)](https://github.com/samuelgursky/davinci-resolve-mcp/releases)
+[![Version](https://img.shields.io/badge/version-2.90.0-blue.svg)](https://github.com/samuelgursky/davinci-resolve-mcp/releases)
 [![npm](https://img.shields.io/npm/v/davinci-resolve-mcp.svg?label=npm&color=CB3837)](https://www.npmjs.com/package/davinci-resolve-mcp)
 [![API Coverage](https://img.shields.io/badge/API%20Coverage-100%25-brightgreen.svg)](docs/reference/api-coverage.md)
 [![Tools](https://img.shields.io/badge/MCP%20Tools-34%20(353%20full)-blue.svg)](#服务器模式)
@@ -12,7 +12,7 @@
 [![Python](https://img.shields.io/badge/python-3.10+-green.svg)](https://www.python.org/downloads/)
 [![License](https://img.shields.io/badge/license-MIT-blue.svg)](https://opensource.org/licenses/MIT)
 
-> 本翻译对应 v2.88.0 版 README。如与英文原版有出入，以 [英文原版](README.md) 为准。
+> 本翻译对应 v2.90.0 版 README。如与英文原版有出入，以 [英文原版](README.md) 为准。
 
 一个 Model Context Protocol (MCP) 服务器，让 AI 助手通过官方脚本 API 控制 DaVinci Resolve Studio（达芬奇）。它提供完整的 API 覆盖，外加带护栏的工作流助手，涵盖剪辑、媒体池整理、渲染设置、审阅标记、调色、Fusion、Fairlight、项目生命周期任务、扩展开发，以及不碰源媒体的媒体分析。
 

package/docs/SKILL.md

@@ -489,7 +489,16 @@ you are on the correct page first.
 Key actions:
 - `launch` — connect to or start Resolve; call this first if any tool returns a
   "Not connected" error
-- `get_version` — returns `{product, version, version_string}`
+- `get_version` — returns `{product, version, version_string, build, mcp}`.
+  `build.unavailable_on_this_build` lists every recorded API surface this build
+  does **not** have; read it before offering anything version-gated. An absence
+  from that list is not a promise a method exists — most of the API has never
+  been version-bisected, so `check_version_support` answers `unknown` for it,
+  and `unknown` means probe with `name in dir(obj)`, never bare `hasattr`
+  (constant `True` on Resolve objects)
+- `check_version_support(symbol?, resolve_version?)` — is one named symbol on
+  this build? Without `symbol`, the same missing-surface list `get_version`
+  carries. No connection needed when `resolve_version` is passed
 - `api_truth(query?)` — look up behaviorally-verified facts about quirky/unreliable
   Resolve API behavior (no connection needed); filter by substring
 - `verification_stats` — readback-verification tally (verified/contradicted/

package/install.py

@@ -36,7 +36,7 @@ from src.utils.update_check import (
 
 # ─── Version ──────────────────────────────────────────────────────────────────
 
-VERSION = "2.88.0"
+VERSION = "2.90.0"
 # Only hard floor: mcp[cli] requires Python 3.10+. There is no upper bound —
 # Resolve's scripting bridge loads into newer interpreters on recent builds
 # (Python 3.14 verified against Resolve Studio 20.3.2). Older Resolve builds

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "davinci-resolve-mcp",
-  "version": "2.88.0",
+  "version": "2.90.0",
   "description": "NPM bootstrapper for the DaVinci Resolve MCP Server.",
   "license": "MIT",
   "author": "Samuel Gursky <samgursky@gmail.com>",