SceneWeaver MCP 0.1.2

SceneWeaver MCP 0.1.2 adds guided client connection and includes the video workflow fixes from PR #1.

Download and install

Extract sceneweaver-mcp-0.1.2-local.zip and run the included installer for your platform. It installs dependencies, detects supported client configuration folders, and offers to connect one selected client. Enter skips registration. Python 3.11/3.12 recommended. Local narration still needs an explicit setup-voice model download.

Client setup

• sceneweaver-mcp clients: read-only detection.
• sceneweaver-mcp connect: choose a detected client interactively.
• sceneweaver-mcp connect --client cursor --dry-run: preview one entry.
• sceneweaver-mcp connect --client cursor: explicitly register it.

Client IDs: codex, cursor, claude-desktop (the last supports Windows/macOS). Registration backs up existing configuration, preserves other servers/settings and file permissions, and refuses malformed, linked, conflicting, or changing files. No broad disk/process scan, background service, automatic restart, approval bypass, or client modification through MCP tools. Plain pip installation never changes client settings. See docs/clients.md for supported paths and privacy considerations.

Fixes

Preserve completed results on cancellation; wait for workers before recording cancellation; finish cleanup before releasing storage; accept valid moving one-second clips; hide render/preview console windows on Windows; exclude generated bytecode from packages.

Validation

52 tests completed on Windows locally and against the installed wheel (one symlink-creation test skipped for account permissions). All six platform/Python CI configurations passed on both the branch and pull request, including package builds and dependency audits. Runtime video tests use real FFmpeg; client tests use temporary configuration files.

Limits

Configuration-folder detection is a hint, not proof a client is installed. Restart the chosen client and approve SceneWeaver there. Only one server can own the default storage; close other connected clients or set separate SCENEWEAVER_HOME paths. Unsupported profiles/formats need manual setup. Voice model setup remains explicit, and stock relevance/rights still require review. This is an early release without independent security certification.

Original code remains Apache-2.0; required third-party notices are preserved. The ZIP includes wheel/source distributions, installers, configuration examples, and documentation. Verify distribution files with SHA256SUMS.

SceneWeaver MCP 0.1.1

SceneWeaver MCP turns your existing assistant's storyboard into a local narrated video. Original code is Apache-2.0; required third-party notices are retained.

Download and run

Download sceneweaver-mcp-0.1.1-local.zip below and extract it. Follow README.md or run the included installer for Windows, macOS, or Linux. Python 3.11 or 3.12 is recommended. The ZIP includes the wheel, source archive, install/run scripts, MCP configuration example, notices, and documentation. SHA256SUMS verifies all three distribution files. Dependencies are downloaded during installation; local voice requires the [voice] extra and an explicit approximately 355 MB model download.

Included

• Storyboard creation, inline previews, media selection, and local media imports.
• Commons and Openverse media search; optional API-Anything bridge with Pexels opt-in.
• Local Kokoro narration and voice preview; original ambient background music with voice ducking.
• Sentence, phrase, or word captions, branding, and duration-controlled 1080p exports.
• Job progress, cancellation, results, asset credits, and quality reports.
• Stdio MCP, managed storage, explicit network consent, public-host restrictions, and bounded workers.

No writing-model API key is required: the connected assistant supplies the script. No upstream MoneyPrinterTurbo engine is bundled.

Validation

All six GitHub jobs passed on Windows/macOS/Linux with Python 3.11/3.12, including tests, packaging checks, and dependency audits. All 34 tests also passed against a clean installed wheel on Windows. A real stdio MCP run produced a 15.0-second 1920×1080 video with local Kokoro narration, captions, and background music. Live API-Anything/Commons search returned six candidates.

Preview limitations

This is an early release, not an independent security certification. Stock relevance and usage rights still need review. Phrase/word timings are estimates within measured sentence boundaries. Pexels availability is fragile; blocked or rate-limited sources are not bypassed. Actual Kokoro voice rendering was exercised on Windows; cross-platform CI uses synthetic speech to keep tests small.

0.1.1 corrects temporary-directory aliases in the render test and upgrades build tooling after the initial source preview.

SceneWeaver MCP 0.1.0 — first public preview

Local MCP video creation using your existing assistant. Apache-2.0 for original code; third-party notices retained.

• Storyboards, native image previews, scene media selection, and local library imports.
• Commons and Openverse media search; optional API-Anything bridge, with Pexels opt-in.
• Local Kokoro narration, voice previews, original background music, captions, and branding.
• Duration-controlled 1080p exports, progress, cancellation, credits, and quality reports.
• Stdio transport, managed files, public-host network restrictions, download consent, and worker limits.

Validation: 34 tests passed both from source and a clean wheel installation on Windows; a real MCP render produced 15.0 seconds at 1920×1080 with local narration and music. Exact-version PyPI vulnerability checks found no published advisories for the 72 installed external packages at release time.

Download the local ZIP, extract it, and follow README.md or the included install scripts. Python 3.11/3.12 recommended. Voice requires the optional [voice] dependencies and an explicit approximately 355 MB model download. No writing-model API key is required.

Early preview: review media relevance and rights before publishing. Word/phrase caption timing is estimated within measured sentence boundaries. Pexels availability is fragile and never bypasses site challenges. macOS/Linux execution remains subject to CI validation.