🎬 UploadCheck MCP

The quality gate between your AI and "publish".

Your agent can generate video. It can't tell you the audio cut out at 0:08, a frame froze at the stitch seam, or the captions are cropped. UploadCheck can — and hands back the exact timestamps to fix.

For anyone whose agent, pipeline, or editor produces media that ships to real people — AI-video pipelines, Shorts/Reels, podcasts, client deliverables.

The problem

An agent that writes code can run the tests. An agent that renders a video has no idea whether it's broken.

So the broken export ships: silence where narration should be, a frozen frame where two clips were stitched, lips out of sync, captions clipped by the platform UI, loudness well off target. Somebody notices in the comments.

UploadCheck is the check your agent calls before it publishes. Point it at the finished file. It returns PASS, WATCH, or BLOCK with per-flag timestamps and machine-actionable fixes — so the agent repairs only the flagged span and re-checks, instead of guessing or shipping.

Install

Terminal

npx -y @drantoniou/uploadcheck-mcp

Claude Code

Terminal

claude mcp add --env UPLOADCHECK_API_BASE_URL=https://api.uploadcheck.app \
  --env UPLOADCHECK_API_KEY=YOUR_KEY \
  --transport stdio --scope user uploadcheck -- npx -y @drantoniou/uploadcheck-mcp

Cursor / Claude Desktop (mcp.json)

config.json

{
  "mcpServers": {
    "uploadcheck": {
      "command": "npx",
      "args": ["-y", "@drantoniou/uploadcheck-mcp"],
      "env": {
        "UPLOADCHECK_API_BASE_URL": "https://api.uploadcheck.app",
        "UPLOADCHECK_API_KEY": "YOUR_KEY"
      }
    }
  }
}

Codex

bash

codex mcp add uploadcheck \
  --env UPLOADCHECK_API_BASE_URL=https://api.uploadcheck.app \
  --env UPLOADCHECK_API_KEY=YOUR_KEY \
  -- npx -y @drantoniou/uploadcheck-mcp

Get a free key at uploadcheck.app — 40+ gates, 200 checked minutes/month, no card. Machine-readable install snippets for every client live in mcp-install.json.

Keep your key private. It's a secret bearer token tied to your plan minutes. Put it in your personal MCP config — never in a committed file, a shared prompt, or a report.

Use it

Once installed, just ask:

text

check ./final-cut.mp4 before I upload

The agent calls qc_run_local_file, polls qc_get_job, then reads qc_get_report and shows you the flags with timestamps.

What it checks

40+ automated gates run on every check, including:


Picture	frozen/looped frames, black frames, canvas fit & pillarboxing, temporal warping, flicker, interlace combing
Audio	dead air, clipping, loudness (LUFS) & true peak, loudness range, channel balance, mains hum, dropouts
Sync & text	lip/audio drift, caption safe-area, text contrast, cropped or overlapping text
Delivery	resolution/codec/frame-rate spec, Shorts/Reels format

Paid plans add AI "oracle" gates — garbled speech, duplicated faces in AI crowds, narration match.

What it doesn't do: it doesn't tell you whether a clip was AI-generated. It assumes it might be, and checks whether it's clean enough to publish.

Tools

Tool	What it does
`qc_estimate_cost`	Estimate checked minutes and effective gates before running
`qc_run_local_file`	Check a local export (reads the file, sends it to the hosted API)
`qc_run_video`	Check a YouTube URL, signed URL, upload id, or base64 payload
`qc_get_job`	Poll status, progress, verdict, metered minutes
`qc_get_report`	The finished report: verdict + timestamped flags + evidence
`qc_get_events`	Lifecycle events, so the agent can explain what ran
`qc_get_artifacts`	List generated report artifacts
`qc_get_marker_csv`	Editor marker CSV (drop flags straight into your NLE timeline)
`qc_create_upload_url`	Signed upload URL for larger files

Pass idempotency_key to qc_run_video when an agent might retry the same asset — the API returns the existing job instead of duplicating the run.

Use profile when you know the media type but don't want to hand-pick gates: auto, shorts, audio, thumbnail, generic_creator_video, and named delivery targets like youtube_shorts, podcast_delivery, broadcast_r128, ai_render_gate. Explicit checks always win.

Environment

server.ts

export UPLOADCHECK_API_BASE_URL="https://api.uploadcheck.app"
export UPLOADCHECK_API_KEY="<workspace_api_key>"

This package is a thin MCP client over the hosted UploadCheck API — checks run server-side, not on your machine. Small and medium local files can be sent inline (base64) without durable storage; use signed upload URLs for large ones.

License

MIT — see LICENSE.

Uploadcheck README