Kinocut 1.15.2: AI Agent Video Editing MCP Server

What changed

Kinocut has released version 1.15.2, an open-source Model Context Protocol (MCP) server, Python library, and kino command-line interface (CLI) designed to provide AI agents with a guardrailed local video editing surface. This latest version introduces several key improvements, including verified sound line and an adversarial-hardened engine. The sound mastering process now preserves stereo through a measured chain, incorporating real caption-speech distance profiles, and the sound-qa-asr feature is designed to fail closed without an actual recognition request. Additionally, four guarded local Revideo operations have been added to the public surface, featuring rendered-format validation. The release-checkpoint function now passes the MCP release gate directly from the CLI. Security has been enhanced through adversarial audit hardening, which includes reporting the actual Kinocut version during MCP handshake and implementing a central guard that rejects outputs aliasing an operation's inputs. Oversized error echoes are also truncated. The total surface area of Kinocut has grown to 201 MCP tools and 173 CLI commands. This release is compatible with mcp-video==1.6.13, which installs kinocut==1.15.2.

Previous versions also brought significant features. Version 1.15.1 was a maintenance release that updated the mcp-video compatibility shim to 1.6.12, republished the legacy registry entry for better metadata pointing, and improved the object-matte engine for streaming decode with bounded memory on large inputs. Version 1.15.0, the Windows/diagnostics release, introduced first-class Windows support and honest diagnostics. This included improved MCP startup failure reporting, a verified MCP server import check via kino doctor, and portable projectstore file locking for Windows compatibility. It also added a windows-latest CI smoke job to guard the platform. A notable feature introduced in 1.15.0 is the 360 dual-cam assembly, allowing stitched equirectangular MP4 files to be reviewed and rendered. Other enhancements in 1.15.0 included faster 360 processing and import paths, lazy public imports to avoid eager loading, and honest ship-seam reporting with default minimum scores for repurposing.

Why it matters for builders

Kinocut 1.15.2 offers AI builders a more reliable and secure platform for integrating video editing into agent workflows. The focus on guardrailed operations and Video Receipts ensures that media edits are traceable and reviewable, reducing the risk of errors or unintended modifications. The adversarial hardening adds a layer of security, making the system more resilient to potential exploits. For developers working with AI agents that need to process and repurpose video content, this release provides a powerful, local-first toolset that bypasses the complexities of direct FFmpeg flag management.

Practical impact

AI builders can leverage Kinocut 1.15.2 to automate tasks such as trimming interviews, adding captions, repurposing content for social media platforms like Shorts and Reels, and performing quality checks before publication. The Python client and kino CLI allow for seamless integration into existing agent frameworks and CI/CD pipelines. Developers can experiment with the new Revideo operations and enhanced sound mastering to improve the quality and consistency of AI-generated video assets. The project's open-source nature and Apache-2.0 license encourage adoption and customization. Builders can install the latest stable version via pip install kinocut or explore development-tip features by cloning the repository.

Caveats and source limits

The source material details the technical specifications and changes in Kinocut versions 1.15.0, 1.15.1, and 1.15.2. However, it does not provide specific benchmark results for the performance improvements or details on pricing, as the software is free and open-source. Information regarding the exact nature of the "adversarial audit" or specific vulnerabilities addressed is not elaborated upon beyond the general hardening measures. The source also mentions a "next public release" but does not specify a date or feature set for it. The exact impact of the adversarial hardening on different types of attacks is not detailed.