AVCLabs Media MCP

The image tools finally shipped. npm @avclabs.ai/media-mcp 0.3.0, published 13 Oct 2026 after five months on 0.2.1, carries all nine tools — create_task, get_task_status and enhance_video_sync for video; enhance_image_sync, colorize_image_sync, denoise_image_sync and get_image_task_status for stills; sam3_predict and get_sam3_task_status for masks — so `npx @latest` now does what the README always said it did. On 17 Oct the repo was split into a monorepo of per-function servers (@avclabs.ai/video-mcp, image-mcp, sam3-mcp) so a client only loads the tool schemas it needs, and the all-in-one package was relabelled 'legacy'. Checked 23 Oct 2026: those three split packages are documented in the README but not on npm yet — the registry 404s all of them — so the command above is still the only one that installs. Expect to switch once they publish; the vendor says the legacy package stays configuration-compatible. The repo still sits at 0 stars. Pricing as rendered 10 Sep 2026: 1 credit per second of video and 1 credit per image mask, against Starter $9.95/mo for 100 credits, Pro $25.95 for 300, Advanced $39.95 for 500, or one-time packs from $19.95/50 — so the entry plan buys about 100 seconds of enhanced footage a month, which makes this a spot-fix tool, not a pass over a reel. MIT, Node 18+, 100MB upload cap, thin client over AVCLabs' HTTP backend at mcp.avc.ai; base URLs are overridable via HTTP_API_BASE_URL and SAM3_API_BASE_URL. Ceiling is 1080p/2K — below Topaz or the Topaz models on fal, so this is the cheap-and-cloud option, not the finishing one. enhance_video_sync blocks and truncates around 45s; use create_task plus get_task_status for anything real. SAM3 segmentation is image-only, so it is a masking tool, not video roto. The vendor documents a JSON mcpServers block rather than a CLI one-liner — the command above is the same command/args/env in claude mcp add form. The generic env var name API_KEY will collide if you set it globally.