Changelog

katto-mcp 0.5.4: list all 16 MCP tools (with input schemas and read-only/destructive annotations) before signing in. Listed on the official Model Context Protocol registry, mcp.so, Smithery and Glama.

A setup page for each client — Claude (Code, Desktop, web), Cursor / VS Code / Windsurf, and ChatGPT — plus a generated 16-tool reference at /mcp/tools kept in sync with the server.

Clip export now runs through a single FFmpeg pipeline instead of two, so most clips render in well under a minute.

In the editor you can set the split ratio for stacked (speaker / gameplay) reframes — 50/50, 60/40, 75/25 and a free slider.

Clip a video straight from your terminal with the `katto` command. The same clipping agent now runs across the app, REST API, CLI and MCP — one shared monthly quota.

Point any OAuth 2.1 MCP client at mcp.katto.tech and sign in with your Katto account — connect Claude, Cursor or ChatGPT with no API key stored on disk.

Submit jobs, list clips, fetch transcripts and usage over HTTP — with idempotency keys, HMAC-signed webhooks and scoped (read/write) API keys. OpenAPI 3.1 spec at /openapi.json.

A "New completed job" trigger and a "Create clip job" action, so Katto plugs into no-code automations.

Dub your clips into 8 languages (English, Spanish, French, Italian, Portuguese, Hindi, Japanese, Chinese).

Transcription and speaker separation now run in a single GPU pass: captions in 99 languages, word-timed in 41, with cleaner multi-speaker splits.

Auto-generated scroll-stopping hook overlays, a publishing calendar with drag-to-reschedule, and a dashboard that charts your clip scores.

Post or schedule clips to YouTube, TikTok, LinkedIn, Instagram, Bluesky, Mastodon and Threads.