Skip to main content

FFmpeg MCP

Run portal-managed FFmpeg tooling for media operations without per-user third-party secrets.

Authentication​

  • Header: Authorization

  • Format: Bearer mad_live_***

  • FFmpeg access is brokered through portal grant headers on the backend.

  • Use URLs or storage locations your FFmpeg connector can access.

Catalog Coverage​

  • Raw provider tools: 55
  • Exposed compatibility tools: 55
  • Agent-ready tools: 55
  • Documented descriptors: 55
  • Public reference pages: 55
  • Source repository: https://github.com/MADPANDA3D/FFMPEG-MCP
  • Catalog version: 2026-07-12.3
  • Catalog verified: 2026-07-12T20:27:56.93004+00:00 (5 days old)
  • Guide verified: not manually verified (not verified)
  • Committed snapshot: catalog-5ec40de147b11c40
  • Docs generated: 2026-07-18T00:19:58.776Z

Read-only Tools​

  • Brand Kit Get — Use this to retrieve one local FFmpeg brand-kit definition by ID. It is read-only and never returns external credentials.
  • Brand Kit List — Use this to list available local brand-kit IDs before selecting one for a render. It is read-only and returns no credential values.
  • Check Configuration — Use this before media work to verify that FFmpeg, storage, queue, signing, and optional export integrations are configured. It reads configuration presence only, contacts no provider, and never returns credential values.
  • Ffmpeg Describe Preset — Use this to inspect one approved encoding preset's container and settings before processing. It is read-only and does not run FFmpeg.
  • Ffmpeg List Presets — Use this to list approved encoding presets before transcode or batch export. It is read-only and does not run FFmpeg.
  • Find Tools — Use this to find FFmpeg tools from a natural-language task. It performs deterministic punctuation-normalized multi-token search over agent-ready descriptors and supports category, risk, tier, and result-limit filters.
  • Get Capabilities — List FFmpeg MCP operations and environment capabilities.
  • Get Endpoint Coverage — Use this to map supported FFmpeg capability areas to native MCP tools and their contract tiers. FFmpeg is a local command-line provider rather than an HTTP API, so this reports capability coverage instead of REST endpoints.
  • Get Tool Usage — Use this after discovery to retrieve one tool's complete descriptor, input and output schemas, safety annotations, aliases, and confirmation rule. It does not execute the selected tool.
  • Job Logs — Use this to read the bounded, sanitized log summary for a previously returned job ID when status alone is insufficient. It never returns request headers.
  • Job Progress — Use this for a compact progress and phase readback for a previously returned job ID. It is read-only and does not retry the job.
  • Job Status — Use this to read the complete status, progress, outputs, QA, and safe error summary for a previously returned job ID. It does not retry or mutate jobs.
  • List Capabilities — Use this to understand the FFmpeg MCP surface by category and risk. Set include_descriptors to true when the caller needs the complete, lossless, versioned ToolManifest for catalog ingestion; it performs no media work.
  • Media Get Download URL — Use this to obtain a time-limited signed download URL for an unexpired staged asset. It does not expose storage credentials or change the asset.
  • Metrics Snapshot — Use this to inspect local FFmpeg queue depth and aggregate runtime counters for operational diagnosis. It is read-only and returns no credentials.
  • Rubric Describe — Use this to inspect one media-quality rubric and its scoring rules before analysis. It is read-only and performs no media work.
  • Rubric List — Use this to list approved media-quality rubrics before analysis or iteration. It is read-only and performs no analysis.
  • Template Describe — Use this to inspect one approved render template's variables and layers before applying it. It is read-only and performs no media processing.
  • Template List — Use this to list the approved server-side render templates before choosing one for template_apply. It is read-only and performs no media processing.

Write and Destructive Tools​

  • Asset Compare — Use this to score and rank multiple staged assets against one named rubric. It queues analysis and returns a job ID or cached ranking without changing the compared media.
  • Audio Duck — Use this to mix voice over music while automatically reducing music during speech. It queues a new audio asset using bounded compressor controls.
  • Audio Fade — Use this to add bounded fade-in and fade-out timing to a staged audio asset. It queues a new output asset rather than modifying the source.
  • Audio Mix — Use this to combine multiple staged audio assets with optional per-track volume, normalization, and duration behavior. It queues one mixed output.
  • Audio Mix With Background — Use this to combine voice and background music with optional ducking and independent gains. It queues a new audio asset and preserves both inputs.
  • Audio Normalize — Use this to create a loudness-normalized audio asset with bounded LUFS, range, peak, format, and bitrate controls. It queues a new output asset.
  • Audio Trim Silence — Use this to remove leading and/or trailing silence from staged audio using a bounded duration and decibel threshold. It queues a new output asset.
  • Batch Export Formats — Use this to render one staged asset into multiple explicitly named approved presets. It queues a bounded batch job and returns its job ID.
  • Batch Export Social Formats — Use this to export one staged asset to configured social formats, optionally overriding the preset list. It queues a bounded batch job.
  • Brand Kit Apply — Use this to apply an existing brand kit and optional text overlay to a staged asset. It queues a new branded output and preserves the source.
  • Brand Kit Delete — Use this only to permanently remove a local brand-kit definition. The Portal must require the exact confirmation phrase before execution; media assets already rendered with the kit are not deleted.
  • Brand Kit Upsert — Use this to create or replace a local FFmpeg brand-kit definition used by later renders. It changes persistent configuration state but does not render media by itself.
  • Campaign Process — Use this to process a bounded set of staged assets across social presets and optional template/brand settings. It queues a campaign batch job.
  • Captions Burn In — Use this to burn SRT, VTT, or timed-word captions into a staged video with bounded brand and safe-zone styling. Supply exactly one caption source; it queues a new rendered asset.
  • Extract Audio — Create an audio-only export from source media.
  • Ffmpeg Thumbnail — Use this to extract a still thumbnail from a staged video at a requested timestamp and optional width. It queues a media job and returns a job ID whose result can be followed with job_status.
  • Ffmpeg Trim — Use this to create a time-bounded clip from a staged asset. It queues a new output asset between start_sec and end_sec; choose reencode=false only when stream-copy boundary accuracy is acceptable.
  • Image To Video — Use this to turn one staged image into a fixed-duration video canvas. It queues a new asset with bounded duration, dimensions, frame rate, and background color.
  • Images To Slideshow — Use this to create a slideshow from ordered staged images using fixed or per-image durations. It queues a new video asset and does not modify inputs.
  • Images To Slideshow Ken Burns — Use this to create a slideshow with bounded Ken Burns pan-and-zoom motion from ordered staged images. It queues a new video asset and preserves inputs.
  • Inspect Media — Extract technical metadata from a media source.
  • Media Export To Discord — Use this to upload a staged asset to a specific Discord channel with an optional message. It writes to an external system and requires configured Discord credentials.
  • Media Export To Drive — Use this to upload a staged asset to the configured Google Drive destination and return its Drive file ID. It writes to an external system and requires Drive configuration.
  • Media Ingest From Drive — Use this to copy a Google Drive file into owned FFmpeg staging storage by file ID and return an asset record. The Drive file must be reachable by the configured ingest path; this reads external content and creates local state.
  • Media Ingest From URL — Use this to copy an allowlisted public media URL into owned FFmpeg staging storage and return a stable asset ID. Do not pass private-network URLs or unsupported content; the server downloads and validates external bytes.
  • Render Iterate — Use this to render, analyze, and boundedly iterate an approved social ad, testimonial, or offer card until its rubric threshold or iteration limit is reached. It queues a non-trivial workflow and never publishes results.
  • Render Offer Card — Use this to produce approved offer-card variants from staged media, headline, price, CTA, brand, captions, and audio inputs. It queues new output assets.
  • Render Social Ad — Use this to produce approved social-ad variants from staged media, copy, captions, brand, and audio inputs. It queues bounded 9:16, 1:1, 4:5, and optional 16:9 outputs.
  • Render Testimonial Clip — Use this to produce approved testimonial variants from staged media, quote, author, captions, brand, and audio inputs. It queues new output assets.
  • Template Apply — Use this to render a staged asset through an approved template with bounded variables and optional brand kit. It queues a new output asset.
  • Transcode Video — Convert source video to target format and codec profile.
  • Video Add Logo — Use this to overlay either a staged logo asset or an allowlisted server logo on a video. It queues a new output asset with bounded position, scale, and opacity controls.
  • Video Add Text — Use this to render a bounded text overlay on a staged video with approved font, color, box, and position controls. It creates a queued output asset and never accepts raw FFmpeg filter expressions.
  • Video Analyze — Use this to analyze technical and optional rubric quality signals for a staged asset. It queues analysis and returns a job ID or cached report; it does not publish or replace the source media.
  • Video Concat — Use this to join staged clips into one ordered video with bounded transition, canvas, and audio controls. It queues a new output asset and preserves the input assets.
  • Workflow Run — Use this to execute a bounded directed workflow of supported FFmpeg operations when a single native tool is insufficient. It accepts only approved node types and queues new output assets.