Built for PBLSH

AIMCPToolsAutomation

Creators upload the clip. AI writes the rest.

PBLSH is a marketplace for UGC video: short, unbranded clips of real people that brands license for their ads. We built the platform, and an AI step that watches every upload and writes its title, description and tags. The library stays searchable, and creators never have to fill in a form.

is all a creator has to upload: Just the fileis all a creator has to upload
title, description and tags written from the clip itself: AI-describedtitle, description and tags written from the clip itself
every suggestion checked before it goes live: Human-approvedevery suggestion checked before it goes live
Claude, ChatGPT or any MCP-ready assistant can do the describing: Any AIClaude, ChatGPT or any MCP-ready assistant can do the describing

The problem

A library is only as good as its tags, and nobody likes tagging

Brands find clips by searching and filtering, so every clip needs a good title, a real description and the right tags. When creators were asked to write these, clips arrived named “IMG_4471”, with an empty description and one tag picked just to get past the form. The worst data ended up exactly where search depends on it.

Before

Creators fill in the form

Titles like “IMG_4471”
Blank descriptions
One tag, chosen to get past the form
Search and filters that can’t find the clip

After

AI writes it, a person approves

The upload form needs only the video
AI watches six frames and writes a clear title and description
Tags come from the library’s own list, so filters keep working
An admin approves, edits or rejects in one click

How it works

From upload to searchable in one approval

Nothing the AI writes goes live on its own. Every suggestion waits in a review queue for a person, which keeps quality in human hands while taking the writing off them.

Creator
Uploads a clip
- Just the video
- Optional note for context
PBLSH + AI
Describes it
- Six stills cut from the clip
- AI reads the frames
- Writes title and description
- Picks tags from the library’s list
Admin
Approves
- Accept in one click
- Edit if needed
- Park or reject
Brands
Find it
- Search
- Filter by tag
- Unlock with credits

A clip's page, with the AI-written description and tags — What the AI wrote for one clip: a title, a detailed description of what happens in the frame, and tags for age, emotion, clothing, setting and niche.

The review queue with AI suggestions waiting for approval — The review queue. Each clip arrives with the AI’s suggestion. Approve, set its credit price, park it or reject it.

The upload form, which only asks for the video — The upload form. “We’ll write the title, description and tags from the clip itself.” Writing them by hand is still there, folded away.

Built for AI

The AI works through PBLSH’s own MCP server

We built PBLSH an MCP server, the open standard AI assistants like Claude and ChatGPT use to connect to software. The assistant asks for the next clip, gets its six frames as images, writes the description, and hands it back. It works the same for creators’ clips and for brands’ own product footage.

Each connection has its own key that can be revoked, and every call is counted, so the team always knows which machine did what. Nothing is hard-wired to one AI model: moving to another means changing the thing that calls the tools, not the platform.

You ask your AI assistant

Suggest metadata for the clips that are waiting.
Describe the new product footage from this morning.
Skip that one, the frames are too dark.

Claude, ChatGPT or any MCP-ready assistant picks the tool, calls pblsh and answers from what comes back, on a laptop or a phone.

Creator clips

suggestions wait for an admin

list_clip_suggestionswhat is waiting to be described
claim_cliptake one clip, with its six frames
suggest_cliphand back title, description, tags
skip_clipput it back, with a reason

Brand footage

descriptions apply directly

list_brand_assetsbrand clips waiting
claim_brand_assettake one, with its frames
describe_brand_assetwrite its title and keywords
skip_brand_assetput it back

MCP connections in the PBLSH console — Connections: each AI client has its own key, its scopes, a count of every call, and a revoke button.

A person has the last word

A creator’s tags only go live when an admin presses approve. The AI suggests; it never publishes.

No master key on a laptop

Each machine gets its own key with only the scopes it needs, instead of a copy of the database password.

Every call metered

The console shows each connection’s calls over time and by tool.

Running the marketplace

Everything else a two-sided marketplace needs

Beyond the AI, PBLSH is a full platform: brands spend credits to unlock clips, creators earn when theirs are used, and the team runs it all from one console.

Library and previews

Watermarked previews are made automatically for every clip, and the full file unlocks with credits.

Request board

Brands post what they need. Creators shoot it, and requests the library already covers are pointed at the clips that exist.

Credits and payouts

Brands spend credits to unlock clips, and creators earn when theirs are used. Creators can also sell a batch outright.

Brand teams

A brand invites its team with owner, editor or viewer seats, and editors download what the owner has unlocked.

Demand analytics

Which tags brands want compared with how many clips the library has, so the team knows what to commission next.

Agreements, automated

When a creator signs the content agreement, our Automator platform emails them the signed PDF and tells the team.

The PBLSH admin console dashboard — The admin console: users, clips waiting for review, and demand at a glance.

Demand analytics comparing requested tags with the library — Demand: what brands are looking for, measured against what the library holds.

Built with

The pieces underneath

Next.js
React
TypeScript
Supabase
Postgres
Cloudflare R2
ffmpeg worker
MCP
Claude
Automator (ours)
Docker
Coolify

The whole platform in English, Bahasa Indonesia and Bahasa Melayu.