Ffmpeg Micro MCP server

What it does

MCP server that lets AI agents create, monitor, and download video transcodes through the FFmpeg Micro REST API. It exposes tools for video transcoding jobs, blueprint workflows (captioning, resizing, watermarking), and file uploads. Intended for developers integrating video processing into AI applications.

• Create a transcode job from a video URL, poll until completion, and retrieve a signed download URL.
• Run a pre-built blueprint workflow like caption-video or product-ad and retrieve the processed output.
• Upload a local video file to FFmpeg Micro storage and use it as input for transcoding or other workflows.

Tools this server exposes

Tool	Does
transcode_video	Create a transcode job from one or more input videos (gs:// or https://) with quality/resolution presets or raw FFmpeg options.
get_transcode	Fetch the current state of a transcode job.
list_transcodes	List transcode jobs with optional filters by status, page, limit, date range.
cancel_transcode	Cancel a pending or processing transcode job.
get_download_url	Generate a 10-minute signed HTTPS URL to download a completed transcode's output file.
transcode_and_wait	Create a transcode job, poll until it finishes, and return the signed download URL in one call.
request_upload_url	Generate a presigned HTTPS URL to upload a local file to FFmpeg Micro storage.
confirm_upload	Complete a file upload and receive the final gs:// URL and probe metadata for use in transcoding.
run_blueprint	Start a blueprint run — a pre-built video workflow such as captioning, resizing, watermarking, or ad insertion.
get_blueprint_run	Fetch a blueprint run's status, current step, and output URLs.
run_blueprint_and_wait	Start a blueprint run and poll until it completes, fails, or pauses for review.
continue_blueprint_run	Resume a paused blueprint run (e.g., caption-video awaiting transcript review) by submitting an approved SRT transcript.

Before you enable it

An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.