Releases

Recent releases across the mcp-tool-shop-org organization. Refreshed 2h ago (1:43 PM). Newest release: 18h ago.

v1.0.0 — Initial Release

  • A SessionStart hook that health-checks your MCP servers, quarantines broken ones, and auto-restores them when they come back online.
  • ```bash
  • pip install mcp-bouncer
  • ```
  • Then add to your Claude Code settings:
  • ```json
  • Real-time teaching intelligence: position tracking, live MIDI feedback, voice filters, seek, and live playback controls.
  • Position tracker** — `PositionTracker` maps time to beat/measure/tempo with binary search for seek operations
  • Live MIDI feedback** — `createLiveMidiFeedbackHook` provides position-aware, measure-level teaching: range warnings, velocity shift detection, density alerts, section boundaries, milestone announcements
  • Voice filters** — `melody-only` (highest note per cluster), `harmony` (lowest), or `all`
  • Seek support** — CLI `--seek N` jumps to second N before playback
  • Pause/resume MCP tool** — `pause_playback` tool pauses or resumes the active playback session
  • Initial public release — MCP server + CLI for AI-powered piano teaching.
  • Session engine** — play/pause/stop with 4 playback modes (full, measure, hands, loop)
  • Speed control** — 0.5x-2x multiplier with per-song tempo override (10-400 BPM)
  • Progress tracking** — configurable callbacks at percentage milestones or per-measure
  • 7 teaching hooks** — console, silent, recording, callback, voice, aside, compose
  • Voice feedback** — VoiceDirective output for mcp-voice-soundboard integration
  • Initial public release — piano sheet music in hybrid JSON + musical-language format, built for LLMs.
  • 10 songs** across 10 genres (classical, jazz, pop, blues, rock, R&B, latin, film, ragtime, new age)
  • Three-layer hybrid format**: metadata + musical language (for LLM reasoning) + code-ready measures (for MIDI playback)
  • Registry API**: `getSong`, `searchSongs`, `getStats`, `validateSong`
  • 34 tests** with full validation coverage
  • Multilingual README (8 languages)

Control Room v1.0.0

  • 1. Download `ControlRoom-win-x64.zip`
  • 2. Extract to a folder
  • 3. Run `ControlRoom.App.exe`
  • See [CHANGELOG.md](https://github.com/mcp-tool-shop-org/control-room/blob/main/CHANGELOG.md) for details.

Soundboard.Client v1.2.0

  • NuGet icon**: Package now displays the Soundboard icon on nuget.org
  • Publishing environment**: Workflow uses NuGet Trusted Publishing with GitHub OIDC
  • No API or behavioral changes from v1.1.0
Mcp Tool Registry v1.0.1
  • Fix: include logo in npm tarball
  • Scoped package with correct org URLs. Unscoped `venvkit` is now deprecated — use `@mcptoolshop/venvkit` instead.
Mcp Tool Registry v1.0.0
  • Stable release — metadata-only registry for MCP Tool Shop tools
  • Python virtual environment diagnostic toolkit — @mcptoolshop scope
  • Initial release — modular control plane for distributed consensus systems
Mcpt v1.0.3
  • Fix PyPI page: absolute logo URL, updated README with mcp-select branding and install instructions.
Mcpt v1.0.2
  • Rename PyPI package from mcpt-cli to mcp-select (mcpt was taken on PyPI). Fixes version mismatch between pyproject.toml and __init__.py.
  • Patch release — version bump for PyPI publish verification.
  • No functional changes from v0.2.0.
  • Turn a ZIP or folder into a guided, LLM-friendly metadata bundle — map, route, and guardrails inside the archive itself.
  • Index comparison** (`zip-meta-map diff`)
  • Compare two `META_ZIP_INDEX.json` files and surface all changes: files added/removed/modified, role changes, plan changes, capability drift, risk flag changes
  • `--exit-code` for CI gating — exit 1 if changes detected
  • `--json` for structured output in pipelines
  • PR comment mode** (GitHub Action)
Mcpt v1.0.1
  • Publish to PyPI as mcpt-cli (mcpt name was taken)
Mcpt v1.0.0
  • First stable release of the MCPT CLI.
  • Full registry integration with `mcp-tool-registry`
  • Tool discovery, search, and installation
  • Workspace configuration via `mcp.yaml`
  • Plugin system for extensibility
  • ```bash
Mcp Tool Registry v0.3.1
  • Publish to npm with automated workflow.
  • Bundles system (core, agents, ops, evaluation)
  • Health reports and facets
  • Site build pipeline
  • Schema validation improvements

v1.0.2 — Stability & Hardening

  • AST-based analysis**: Complexity scoring and injection detection now use real AST walks instead of string counting/regex
  • Deterministic builds**: `uv.lock` committed, CI uses `uv sync --frozen`, lock-drift check prevents silent dep changes
  • Build smoke test**: Every push builds a wheel, installs in a clean env, and verifies imports work
  • Repo governance**: CODEOWNERS, issue/PR templates, Dependabot (pip + GitHub Actions, weekly)
  • Single-source versioning**: `__version__` reads from `importlib.metadata`, test enforces parity with `pyproject.toml`
  • `confidence` field renamed to `signal_strength` on all suggestion/issue dataclasses — **backward compatible**: `to_dict()` still emits both keys, `@property` aliases preserved
  • `categorize_tool()` now falls back to description matching when name has no category keywords
  • `analytics.last_success_at` stores real timestamps instead of the literal string "CURRENT_TIMESTAMP"
  • `SyncEmbedder` no longer crashes when called from inside a running event loop (Gradio, FastMCP)
  • UI `run_async` helper replaced with deterministic loop detection (no more `RuntimeError` roulette)
  • Removed discarded `sequence_hash` read in chain detection
  • Private `_backends` dict access replaced with public `is_backend_connected()` API

v1.0.0 - Streaming Prosody Engine

  • All notable changes to this project will be documented in this file.
  • Deterministic Prosody Engine**: Core DSP logic for expressivity control.
  • Micro-Timing Control**: Sub-millisecond precision for rhythm and duration.
  • Streaming Architecture**: Low-latency audio processing pipeline.
  • Reference Handbook**: Comprehensive guide to vocology and prosody concepts.
  • `voice-engine-dsp`: Primary DSP package.

v2.5.0 — MCP Integration & Agent Interoperability

  • Voice Soundboard v2.5.0 — the agent bridge release.
  • MCP Server** — embedded MCP-compliant server with tool registration, WebSocket transport, and configurable concurrency
  • Core Tools** — `voice.speak`, `voice.stream`, `voice.interrupt`, `voice.list_voices`, `voice.status`
  • Agent Sessions** — session-scoped synthesis with agent ownership and priority-based interruption
  • Streaming Interrupts** — structured interrupt handling with fade-out and rollback points
  • Observability** — synthesis metadata with latency, duration, cost estimates, and cache hit tracking

v1.0.0 — Security Hardening & Release Ready

  • The soundboard plugin hits **v1.0.0** — fully hardened, documented, and ready for production use.
  • | Tag | Phase | Highlights |
  • |-----|-------|------------|
  • | v0.1.0 | Core speech pipeline | SSML-lite, smart chunking, multi-chunk synthesis |
  • | v0.2.0 | Multi-speaker dialogue | Script parser, auto-casting, stage directions |
  • | v0.3.0 | Emotion & ambient | 8 emotions, SFX tags, inner monologue |
  • A token-based lease governor for AI calls — small enough to embed anywhere, strict enough to prevent stampedes.
  • Concurrency control** — weighted slots with interactive reserve
  • Rate limiting** — requests/min and tokens/min with rolling windows
  • Fairness** — prevent actor monopolization with soft caps
  • Adaptive tuning** — auto-adjust concurrency from deny rate and latency
  • Leases** — acquire before, release after, auto-expire on timeout

v0.4.0-alpha.0 — Optional Adapters

  • Five adapters for common AI integration patterns, all living in separate subpath exports for tree-shaking.
  • | Adapter | Import | Description |
  • |---------|--------|-------------|
  • | **fetch** | `throttleai/adapters/fetch` | Wraps any `fetch`-compatible function |
  • | **OpenAI** | `throttleai/adapters/openai` | Wraps `openai.chat.completions.create` with token estimation |
  • | **Tool** | `throttleai/adapters/tools` | Wraps any async function (embeddings, rerankers, etc.) |

v0.3.0-alpha.0 — Usability & DX

  • Presets** — `presets.quiet()`, `presets.balanced()`, `presets.aggressive()` return opinionated `GovernorConfig` objects. Spreadable and overridable.
  • `snapshot()`** — `gov.snapshot()` returns a read-only view of current state (concurrency, rate pools, flags). Print a one-line health status from any monitoring loop.
  • `onEvent` hook** — Optional `GovernorEventHandler` fires structured events on `acquire` / `deny` / `release` / `expire` / `warn`. No logging by default — you decide.
  • Better deny guidance** — `recommendation` strings now include concrete numbers (slot counts, rate usage, actor IDs, retry timing).
  • `waitForRetry()` + `retryAcquire()`** — Retry helpers: `waitForRetry(decision)` sleeps for `retryAfterMs`, `retryAcquire(gov, req, { maxAttempts })` wraps acquire with automatic retry.
  • Strict mode** — `strict: true` catches dev mistakes: double release → error, unknown leaseId → error, long-held leases (>80% TTL) → warn event.

v0.2.0-alpha.0 — Phase 2: Smarter Decisions

  • 5 commits that make the governor's decisions more intelligent without changing the API surface or adding runtime dependencies.
  • Better retryAfter** — computed, not heuristic
  • Concurrency denial retryAfterMs derived from earliest lease expiry
  • Rate denial retryAfterMs computed from rolling window math
  • All values clamped to [25ms, 5000ms] via `clampRetry()`
  • Weighted concurrency** — `estimate.weight`

LeaseGate Lite v0.1.0

  • Smooths local AI workload bursts so your PC stays responsive during heavy usage.
  • Applies practical throughput controls (concurrency, adaptive throttling, rate limits) with sensible defaults.
  • Gives you clear, auditable visibility through live status, event trail, diagnostics export, and in-app audit harness.
  • It does **not** upload prompts or call external telemetry services by default.
  • It does **not** act as cloud orchestration or multi-host fleet management.
  • It does **not** replace your model runtime; it only governs local request pressure and pacing.

v0.1.0 — Security Hardening & Reliability

  • LeaseGate v0.1.0 delivers defense-in-depth security hardening and reliability improvements across the full governance stack. All 5 implementation phases are complete, tested, and production-ready.
  • Command injection prevention** — Shell metacharacter blocklist and direct process execution (no `cmd.exe /c` indirection) in tool isolation
  • Service account token hashing** — SHA-256 hash comparison by default with plaintext backward compatibility
  • Payload size limits** — 16 MB cap on named pipe message framing
  • Path traversal protection** — All file export endpoints reject `..` traversal and constrain to temp/app data directories
  • CSV formula injection prevention** — Report exports escape formula injection characters (`=`, `+`, `-`, `@`)

InControl-Desktop v1.2.0

  • 1. Download the MSIX package below
  • 2. Verify the SHA256 checksum
  • 3. Double-click to install
  • ```powershell
  • (Get-FileHash InControl-Desktop-1.2.0-unsigned.msix -Algorithm SHA256).Hash
  • ```

InControl-Desktop v1.1.0

  • 1. Download the MSIX package below
  • 2. Verify the SHA256 checksum
  • 3. Double-click to install
  • ```powershell
  • (Get-FileHash InControl-Desktop-1.1.0-unsigned.msix -Algorithm SHA256).Hash
  • ```

Soundboard.Client v1.1.0

  • Multi-target:** net8.0, net9.0, net10.0
  • Fix:** EngineInfo deserializes snake_case JSON from real engines (JsonPropertyName)
  • Fix:** WsUri uses UriBuilder — https→wss works correctly
  • Fix:** SpeakRequest.ResolvedRequestId stable per instance (lazy)
  • Fix:** SpeakAsync gracefully closes WebSocket in finally block
  • Perf:** ConfigureAwait(false) on all internal awaits

v1.0.0 — Shared Content System Core

  • Portable, local-first content ingestion + normalization + indexing for **Dev-Op-Typer** (Windows) and **linux-dev-typer** (Linux).
  • Deterministic language detection** — 3-tier: user choice > file extension > heuristic fallback
  • Normalization** — UTF-8 (lossy fallback), LF-only line endings, append trailing newline
  • Metrics** — `lines`, `characters`, `symbolDensity` (29-char frozen set incl `"` and `'`), `maxIndentDepth`
  • Content-addressed IDs** — SHA-256(language + normalized code), stable across platforms
  • Deduplicated, deterministically ordered index** — by language then title

v1.2.0 - Enhanced Vortex Rendering

  • Catmull-Rom Spline Smoothing**
  • Trajectory paths now render with smooth, organic curves instead of jagged line segments. Each segment is interpolated with 8 subdivisions for fluid motion.
  • Adaptive Stroke Width**
  • Line thickness now varies inversely with velocity - faster motion creates thinner strokes, slower motion creates thicker strokes. Like calligraphy, this adds natural visual rhythm to the trajectory.
  • Trail Opacity Decay**
  • Older trajectory points fade naturally, with recent path emphasized. The fade starts at 40% opacity and increases toward the current position, making it easy to see temporal progression.

v1.1.0 - Zero-Warning Build & Complete Crash Recovery

  • Session Recovery**: Full crash recovery now restores playback position, current page, and playback state
  • Automatically reloads the file that was open when the app crashed
  • Resumes playback from exact position
  • Navigates to the page you were viewing
  • Zero-Warning Build**: Eliminated all 11 compiler warnings
  • CS0108: Added \

Soundboard SDK v1.0.0

  • The first stable release of the Soundboard SDK: a standalone .NET 8 library for streaming TTS from any engine implementing the Soundboard API contract.
  • `ISoundboardClient` interface: health, presets, voices, speak, stop
  • Streaming audio via `IProgress<AudioChunk>` over WebSocket
  • `CancellationToken` on every async method
  • `SoundboardClientOptions` with `SOUNDBOARD_BASE_URL` env var support
  • Runtime API version check with warning on mismatch