* ci: speed up PR deploys by decoupling from standard image build

Extract shared docker build logic into a reusable workflow
(docker-build-image.yml) to eliminate duplication.

- build-cloud and build-standard now call the same reusable workflow
- deploy-pr only waits for build-cloud (~3min vs ~12min)
- Standard image still builds in parallel for releases

Before: deploy-pr waited ~12 min (both images)
After: deploy-pr waits ~3 min (cloud only)
Signed-off-by: livepeer-robot <[email protected]>

* fix: set permissions at workflow level for reusable workflow

Reusable workflows inherit GITHUB_TOKEN permissions from the caller
workflow, not from their own job-level permissions block.

Move permissions to workflow level so the reusable workflow can
request packages:write for pushing Docker images.

Signed-off-by: livepeer-robot <[email protected]>

* preserve comments

Signed-off-by: Max Holland <[email protected]>

---------

Signed-off-by: livepeer-robot <[email protected]>
Signed-off-by: Max Holland <[email protected]>
Co-authored-by: livepeer-robot <[email protected]>
Co-authored-by: Max Holland <[email protected]>

Signed-off-by: Max Holland <[email protected]>

Signed-off-by: Max Holland <[email protected]>

Prefix logs with remote inference connection ID

Change disconnected state text from "Disconnected" to "Enable Remote
Inference" so new users understand the feature is available. Bump opacity
on all three header icons (cloud, plugins, settings) for better visibility.
Update e2e test selector to match new title.

Signed-off-by: emranemran <[email protected]>

feat: improve cloud button discoverability and header icon visibility

* Use daydream API to check if plugin is allowed

Signed-off-by: Max Holland <[email protected]>

* address review comment, return 503 on daydream API error

Signed-off-by: Max Holland <[email protected]>

---------

Signed-off-by: Max Holland <[email protected]>

This means we can avoid waiting for a restart when we clear up plugins on remote inference disconnect

Signed-off-by: Max Holland <[email protected]>

Signed-off-by: Max Holland <[email protected]>

Signed-off-by: Thom Shutt <[email protected]>

* Add MCP server for AI-assisted pipeline control

Expose Scope's API as MCP (Model Context Protocol) tools so AI assistants
can programmatically manage pipelines, control parameters, capture frames,
and interact with a running Scope instance over stdio.

Backend:
- Add mcp_server.py: thin MCP-over-stdio client that proxies to the Scope
  HTTP API, with tools for pipeline lifecycle, parameter control, frame
  capture, session metrics, recording, LoRA/plugin management, and more
- Add REST endpoints for headless sessions (start/stop stream, get/set
  parameters, capture frame, session metrics, unload pipeline, recording)
- Add HeadlessSession to WebRTCManager for sessions without a browser
- Add broadcast_notification() for pushing parameter changes to frontends
- Add VideoProcessingTrack.get_last_frame() public accessor
- Defer logging setup to run_server() so the MCP subprocess does not
  create a competing log file that shadows the server's logs
- Add --mcp CLI flag and optional `mcp` dependency group

Frontend:
- Add onParametersUpdated callback to useUnifiedWebRTC for external
  parameter updates (REST API, MCP, OSC)
- Extract applyBackendParamsToSettings in StreamPage so both local sends
  and external pushes share the same settings-sync path

Signed-off-by: RyanOnTheInside <[email protected]>

* review fixes

Signed-off-by: RyanOnTheInside <[email protected]>

* base64 fix

Signed-off-by: RyanOnTheInside <[email protected]>

* Remove recording and unload_pipeline MCP tools

These endpoints are not needed for MCP workflows - removes start_recording,
stop_recording, and unload_pipeline tools along with their references in
instructions and docstrings.

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* Rename session_router to mcp_router and simplify session endpoints

Replaces session_id path parameters with implicit active-session lookup
for stop, recording start/stop endpoints, aligning the REST API with
MCP server usage patterns.

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* Simplify headless session to single-session model and remove unused endpoints

Remove recording and pipeline unload endpoints from MCP router, and
refactor WebRTCManager from a multi-session dict to a single headless
session slot since only one headless session is supported at a time.

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafal Leszko <[email protected]>

* Extract HeadlessSession from webrtc.py into its own module

HeadlessSession has no WebRTC dependencies — it runs FrameProcessor
directly. Moving it to headless_session.py clarifies the separation
between WebRTC and MCP-only session management.

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafal Leszko <[email protected]>

* Rename headless_session.py to headless.py

Consistent with sibling module naming (webrtc.py, recording.py).

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafal Leszko <[email protected]>

* add v2v mode

Signed-off-by: RyanOnTheInside <[email protected]>

---------

Signed-off-by: RyanOnTheInside <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>
Signed-off-by: Rafal Leszko <[email protected]>
Co-authored-by: Rafał Leszko <[email protected]>
Co-authored-by: Claude Opus 4.6 <[email protected]>

* deps and desc

Signed-off-by: RyanOnTheInside <[email protected]>

* port flexibility, ocp

now you tell claude which port to connect on, eg: "connect to scope on port 8009". params are dynamic.

Signed-off-by: RyanOnTheInside <[email protected]>

* Move mcp from optional extra to core dependency

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

---------

Signed-off-by: RyanOnTheInside <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>
Co-authored-by: Rafał Leszko <[email protected]>
Co-authored-by: Claude Opus 4.6 <[email protected]>

* Add graph mode backend: graph executor, schema, state, and pipeline changes

Introduce server-side graph execution infrastructure including graph schema
definitions, graph executor for processing node graphs, and graph state
management. Update pipeline manager, frame processor, and pipeline processor
to support graph-based execution. Add graph inputs/outputs to pipeline schemas.

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* Refactor and harden graph mode backend

Squashed follow-up fixes and improvements:
- Fix graph executor pipeline lookup to use node ID instead of pipeline ID
- Fix review issues in pipeline processor and frame processor
- Validate graph edge ports and clean up _load_events on unload
- Acquire input_queue_lock when resizing graph queues
- Remove pipeline throttler and add graph validation
- Move VACE input video routing from runtime to graph construction
- Remove redundant video-only normalization from pipeline_processor
- Reject duplicate stream edges targeting the same input port
- Fix unknown node_id falling back to broadcast in update_parameters
- Simplify graph_executor.py build_graph wiring
- Remove unused graph_state.py module
- Remove old pipeline chaining remnants
- Move build_linear_graph from frame_processor.py to graph_schema.py
- Fix queue resize
- Refactor pipeline reconciliation logic for readability and add unit tests
- Fix ruff B905 lint error: add strict=True to zip() call

Signed-off-by: Rafał Leszko <[email protected]>

* Fix extra params filtering to use output queues instead of hardcoded key

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

---------

Signed-off-by: Rafał Leszko <[email protected]>
Co-authored-by: Claude Opus 4.6 <[email protected]>

Signed-off-by: Rafał Leszko <[email protected]>

* DMX input

Signed-off-by: Thom Shutt <[email protected]>

* Code review fixes

Signed-off-by: Thom Shutt <[email protected]>

* Revert logging changes

Signed-off-by: Thom Shutt <[email protected]>

* fix(settings): dedupe DMX param options by key in select dropdown

Prevent same param from multiple pipelines appearing multiple times
(e.g. 'zoomzoomzoom' in trigger) by tracking seen keys when grouping.

Signed-off-by: Rafał Leszko <[email protected]>
Made-with: Cursor

* fix(settings): mark form dirty on logging toggle and guard port apply on listening state

handleToggleLogging now sets dirty so the logging change is persistable.
handleApplyPort checks updated.listening before clearing dirty/toasting
success; shows an error toast if the server didn't start listening.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* fix(dmx): validate port range, normalize mappings, and fix channel indexing

Adds preferred_port bounds check, normalizes mappings through
mappings_to_dict before saving, and corrects off-by-one in unmapped
channel logging to use 1-based indexing.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* feat(dmx): make DMX input opt-in (disabled by default)

DMX server was always started and the frontend always opened an SSE
connection, even when users don't use Art-Net. Now DMX is disabled by
default and can be enabled via a toggle in Settings > DMX. The server
only binds the UDP port and the frontend only opens the SSE stream
when DMX is explicitly enabled.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* feat(dmx): add debug logging for incoming Art-Net packets

Log the first UDP packet unconditionally to confirm connectivity, and
log every packet (with source address) when verbose logging is enabled.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* fix(dmx): enforce consistent 1-based channel indexing and port validation

Use Pydantic Field(ge=1024, le=65535) on all DMX port fields instead of
ad-hoc manual checks. Align frontend channel inputs to 1-512 to match
the server 1-based convention. Remove misleading dirty flag from the
logging toggle since it already persists server-side.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* Switch to debug logs

---------

Signed-off-by: Thom Shutt <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>
Co-authored-by: Rafał Leszko <[email protected]>
Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>

The input source loop throttled frame reception based on pipeline FPS,
creating a feedback loop: sleep(1/FPS) + receive_latency caused the
effective input rate to always undershoot the target, driving FPS
downward. For sources with non-trivial receive overhead (NDI, Syphon)
this spiraled to ~5.6fps (v0.1.6) or ~2.8fps (v0.1.7). The v0.1.7
regression was caused by the batch-level FPS tracking (66e8d31) being
more stable — the old per-frame jitter partially counteracted the
spiral via temporary FPS overshoots.

Now the loop receives as fast as the source provides, matching the
WebRTC camera input path. Backpressure is handled by put_nowait()
dropping frames when the queue is full.

Signed-off-by: Rafał Leszko <[email protected]>
Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>

* Tempo sync api with bpm, beat_phase, bar_position, beat_count, and is_playing.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Small refactor.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Tempo sync api with bpm, beat_phase, bar_position, beat_count, and is_playing.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Small refactor.

Signed-off-by: BuffMcBigHuge <[email protected]>

* feat: beatsync with timed params

Signed-off-by: RyanOnTheInside <[email protected]>
Signed-off-by: BuffMcBigHuge <[email protected]>

* Modulation engine and pipeline.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Development and experimentation of pipeline modulation with prompts, steps and others.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Linting.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Missing tempo sync panel.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Signing, merges, repair from bad rebase.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Small fix.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Fix small rebase issue.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Update to tempo sync document.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Code cleanup and refactor.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Linting/formatting.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Better handling.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Add libasound2 to Dockerfiles

* Update doc and CLAUDE.md

* Fix tempo sync docs: use uv sync --extra and remove libasound2 from Dockerfiles

MIDI requires local hardware access so libasound2 is not useful in containers.
Also corrects uv sync --group to --extra across docs.

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafal Leszko <[email protected]>

* Revert unnecessary DownloadDialog prop changes from tempo sync PR

Remove isCloudMode and onOpenLoRAsSettings props that were added to
DownloadDialog solely to suppress TypeScript errors in StreamPage,
but were never actually used by the component.

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* Fix quantize detection missing non-primitive discrete params

transition (object) and denoising_step_list (array) were not caught by
the typeof === "boolean" / "string" checks, so they bypassed beat
quantization. Use a DISCRETE_PARAM_KEYS allowlist for key-based
detection of known discrete parameters.

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* Revert unintended package-lock.json changes

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* Extract tempo sync endpoints into dedicated router

Moves the 5 tempo endpoints from app.py into tempo_router.py,
following the same APIRouter pattern as mcp_router.py.

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* Remove redundant mcp optional dependency group

mcp>=1.0.0 is already in core dependencies, so the optional-dependencies
entry was unnecessary.

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* Improve TempoSource/TempoSync robustness and consistency

Make TempoSource an ABC with abstract methods, add _enabled_lock for
thread-safe access, prune dead notification senders, unify install hints
to use `uv sync --extra`, and clean up redundant .get() fallbacks.

Co-Authored-By: Claude Opus 4.6 <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* Update uv.lock

Signed-off-by: Rafał Leszko <[email protected]>

* Refactor tempo sync logic in PipelineProcessor

Extract beat boundary calculation to a standalone get_beat_boundary()
in tempo_sync.py and consolidate inline beat state handling into a
dedicated _apply_tempo_sync() method.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* Fix review issues: lock consistency, encapsulation, logging, and lint warnings

- Use `self.enabled` property (with lock) in `get_status()` instead of bare `self._enabled`
- Add `PipelineProcessor.set_beat_cache_reset_rate()` to encapsulate private attribute access
- Downgrade noisy per-frame "Extra params" log from info to debug
- Convert f-string logger calls to %-style lazy formatting across tempo sync modules
- Fix ESLint exhaustive-deps warnings in StreamPage prompt cycle effect

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* Removal of demo pipelines.

Signed-off-by: BuffMcBigHuge <[email protected]>

* Fix to clamping of of modulation result, cache state invalidation.

Signed-off-by: BuffMcBigHuge <[email protected]>

* fix: restore OSC SSE subscription and dmx_server global declaration

The OSC SSE EventSource subscription was accidentally deleted when the
tempo sync useEffect was added, breaking OSC parameter control. The
dmx_server variable was also dropped from the global declaration in
lifespan(), causing get_dmx_server() to always return None.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* fix: seed overflow, step list descending invariant, and tempo sync lock race

- Use 2**31-1 as upper bound for random seed to stay within signed 32-bit range
- Add forward pass in modulate_step_list to fix strictly descending invariant
  when values are clamped to hi; fall back to clamped original if range is
  too narrow for distinct values
- Unify _source_lock and _enabled_lock into a single _state_lock in TempoSync
  so enable/disable/get_beat_state see consistent _source + _enabled state

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* fix: guard mido import and fix scheduler quantize_mode race condition

Wrap `import mido` in try/except like the Link adapter does, preventing
ImportError when the module is imported without mido installed. Also move
_quantize_mode and _lookahead_ms reads/writes under the scheduler lock
to eliminate a TOCTOU race where the mode could change between the
initial check and delay computation.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* fix: make left sidebar sections scroll together as one unit

Move overflow scrolling from InputAndControlsPanel to the parent
container so Tempo Sync and Outputs scroll with Input & Controls
instead of pushing it off-screen.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* feat: include Ableton Link in Electron app builds

Add --extra link to uv sync in all three platform setup paths so
aalink is installed automatically in the desktop app.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

* fix: include --extra link in uv run so aalink survives implicit sync

uv run performs its own sync before executing, which can remove aalink
installed during setup. Pass --extra link to uv run as well.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>

---------

Signed-off-by: BuffMcBigHuge <[email protected]>
Signed-off-by: RyanOnTheInside <[email protected]>
Signed-off-by: Rafał Leszko <[email protected]>
Signed-off-by: Rafal Leszko <[email protected]>
Co-authored-by: RyanOnTheInside <[email protected]>
Co-authored-by: Rafal Leszko <[email protected]>
Co-authored-by: Claude Opus 4.6 <[email protected]>

Signed-off-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: leszko <[email protected]>