Uploadcheck

What the uploadcheck MCP server does

The uploadcheck MCP server checks finished media before it is published. It is intended for AI video workflows, short-form content, podcasts, client deliverables, and other pipelines where an exported file needs a quality gate before delivery. A check produces a PASS, WATCH, or BLOCK verdict, along with timestamped flags, evidence, and repair-oriented information that an agent can use to target a specific section rather than regenerate the whole asset.

The service covers more than 40 automated gates. Picture checks include frozen or repeated frames, black frames, canvas fit, pillarboxing, temporal warping, flicker, and interlace combing. Audio checks include dead air, clipping, loudness, true peak, loudness range, channel balance, mains hum, and dropouts. Other checks cover lip and audio drift, caption safe areas, text contrast, cropped or overlapping text, and delivery requirements such as resolution, codec, frame rate, and Shorts or Reels formatting. Paid plans add AI-based checks for garbled speech, duplicated faces in AI crowds, and narration matching.

How it works

The package is a thin MCP client for the hosted UploadCheck API. Checks run remotely rather than on the local machine. Small and medium local files can be sent inline as base64; larger files can use a signed upload URL created with qc_create_upload_url.

A typical flow estimates cost, starts a check, polls its job, and retrieves the completed report. qc_get_events provides lifecycle information so an agent can describe what ran. Reports can also expose generated artifacts and an editor marker CSV for importing flagged timestamps into an NLE. Passing an idempotency_key to qc_run_video lets retries return the existing job instead of creating a duplicate run.

Profiles such as auto, shorts, audio, thumbnail, generic_creator_video, youtube_shorts, podcast_delivery, broadcast_r128, and ai_render_gate select media-oriented gate sets. Explicit checks take precedence when supplied.

Setup and configuration

Install the package with:

npx -y @drantoniou/uploadcheck-mcp

Configure UPLOADCHECK_API_BASE_URL with https://api.uploadcheck.app and provide UPLOADCHECK_API_KEY with a workspace API key. The key is a bearer secret tied to plan minutes, so it should remain in a personal MCP configuration and not in committed files, prompts, or reports. A free key is available from uploadcheck.app and includes 200 checked minutes per month according to the README.

The README documents stdio setup for Claude Code, Cursor, Claude Desktop, and Codex. The package can be added with each client’s MCP configuration or command-line registration method.

Tools and capabilities

• qc_estimate_cost estimates checked minutes and effective gates before execution.
• qc_run_local_file reads and submits a local export.
• qc_run_video accepts a YouTube URL, signed URL, upload ID, or base64 payload.
• qc_get_job returns status, progress, verdict, and metered minutes.
• qc_get_report returns the completed verdict, timestamped flags, and evidence.
• qc_get_events returns lifecycle events.
• qc_get_artifacts lists generated report artifacts.
• qc_get_marker_csv returns editor marker data.
• qc_create_upload_url creates a signed upload URL for larger files.

Limitations and notes

The uploadcheck MCP server does not determine whether a clip was AI-generated; it checks whether the media is clean enough to publish. API processing is hosted, so the media is sent to the service rather than checked entirely on the local machine. Plan minutes and paid-plan availability affect usage of the service and its optional AI gates.