Popcraft

io.github.popcraft-aiPublished by GitHub account popcraft-aiv1.0.1Updated Oct 24, 2026

AI video, images, music & SFX: Seedance 2.5, Veo 3.1, Kling 3.0, Nano Banana Pro, 20+ models.

VerifiedStreamable HTTPWeb executableMedia & DesignAI & ML

Overview

AI-generated overview

Popcraft lets an assistant generate and enhance AI video, images, audio and 3D through 20+ hosted models, with credit-cost previews and account tools.

What you can do
  • Generate video, images, audio (TTS, sound effects, music) and 3D from text or references, and list voices
  • Upscale images and video, re-render drafts at 1080p, and post-process 3D
  • Upload local media via a signed shell recipe or import public URLs, then browse the media library and reusable characters and styles
  • Explore model capabilities and credit costs, display self-refreshing job cards, revisit past generations and file results into projects
  • Load production playbooks (UGC ad, thumbnail, product ad, voiceover, model-specific prompting) and check balance, plans and transactions
When to use it
Useful when you want an assistant to produce marketing or social media assets — product teasers, UGC ads, thumbnails, brand image sets, voiceovers — or to upscale and chain existing media, without leaving the chat.
Requirements
A remote streamable HTTP endpoint with OAuth 2.1 sign-in; the first tool call opens a browser sign-in where you press Allow, after which tokens refresh automatically. A Popcraft account is required; new accounts get 100 free credits. No API keys or environment variables are declared.
Before you install
Generations spend credits from your Popcraft account; use the get_cost preview before submitting, since some premium models are paid-plan only. Uploads send your reference media to Popcraft, and results are stored in your account history. Reference media for the Seedance 2.0 family goes through real-person verification and may be rejected. Jobs are asynchronous and can take minutes.

Installation

In SourceWeft

  1. Open Popcraft in the dashboard and add it to a workspace.
  2. Connect the required credentials. They are stored privately and never shown on public pages.
  3. Enable the server for the chats that should use its tools.

Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.

Use with an agent

Paste this prompt into Claude Code, Codex, Cursor or another agent — or into SourceWeft chat. The agent reads this server's install guide, shows you its publisher, endpoint and sign-in, and adds it to your client once you agree. Keys never go into a command.

Read https://sourceweft.com/mcp/io.github.popcraft-ai%2Fpopcraft-mcp/install.md and connect the MCP server it describes to my client. Before changing any config, show me its publisher, endpoint and sign-in method and wait for my OK. Never put an API key in a command or in a file I would commit; ask me where it should come from.

Read the install guide the agent follows

Other MCP clients

Add this to your client's mcpServers config. The client completes OAuth sign-in on first connect.

{
  "mcpServers": {
    "popcraft": {
      "type": "http",
      "url": "https://popcraft.ai/api/mcp"
    }
  }
}

README

[Popcraft MCP — generate video, images, music and sound effects from any AI agent. One endpoint, 20+ frontier models, no API key. https://popcraft.ai/api/mcp]

[MCP · Streamable HTTP] [Auth · OAuth 2.1] [Models · 20+] [License · MIT]

Docs & one-click setup · Models · Pricing · Recipes

https://popcraft.ai/api/mcp

[Packshot, lifestyle scene, campaign poster and product loop — all generated through the connector]

One product photo in, a whole listing out — packshot, lifestyle scene, campaign poster and product loop, generated through the connector in recipe 04.


Quick start

Every client connects the same endpoint; only the "add server" step differs. The first tool call opens the Popcraft sign-in in your browser — press Allow once and the connection stays live with automatic token refresh. New accounts get 100 free credits.

ClientSetupGuide
Claude (web / desktop)Settings → Connectors → Add custom connector → name Popcraft → paste the URL → Connect → Allowclients/claude.md
Claude Codeclaude mcp add --transport http popcraft https://popcraft.ai/api/mcp then /mcp → popcraft → Authenticateclients/claude-code.md
CursorAdd to ~/.cursor/mcp.json (below), enable popcraft in MCP settings, sign in on first useclients/cursor.md
ChatGPTSettings → Apps & Connectors → Advanced → Developer mode on → Create → name Popcraft → paste the URLclients/chatgpt.md
CodexPaste: "Help me connect the Popcraft MCP at https://popcraft.ai/api/mcp — add it as a streamable HTTP server named popcraft, then walk me through the sign-in."clients/codex.md
Anything else (OpenClaw, Hermes, your own agent)Streamable HTTP + OAuth 2.1 with discovery metadata and dynamic client registration — the URL alone is enoughclients/other.md
Cursor config snippet
json
{  "mcpServers": {    "popcraft": {      "url": "https://popcraft.ai/api/mcp"    }  }}

Try it

Paste any of these into a fresh chat once the connector is on:

Use the Popcraft MCP to make a 10-second cinematic product teaser — preview the credit cost first.

Make me a 15-second vertical UGC ad for my product on Popcraft — hook people in the first second. (loads the ugc-ad skill)

I need a YouTube thumbnail for my video about why cheap mechanical keyboards beat expensive ones — make it pop, using Popcraft. (loads the thumbnail-pro skill)

Shoot a brand image set for the citrus soda — hero can, lifestyle scene, pattern tile — then voice a 15-second tagline.


What your agent gets

GroupToolsWhat they do
Createpopcraft_generate_video
popcraft_generate_image
popcraft_generate_audio
popcraft_generate_3d
popcraft_effects_show
popcraft_list_voices
Text-to-video, image-to-video, first/last frame, reference- and audio-driven video
Every frontier image model, edits with references
TTS, sound effects, music
3D generation
Curated video effects
Playable voice samples
Enhancepopcraft_upscale_image
popcraft_upscale_video
popcraft_upgrade_video
popcraft_postprocess_3d
Topaz 2×/4×, nine modes
2×/4× with fps control
Re-render a Seedance 2.5 draft at 1080p
3D post-processing
Media & referencespopcraft_media_upload
popcraft_media_import_url
popcraft_media_upload_widget
popcraft_media_upload_inline
popcraft_media_confirm
popcraft_show_medias
popcraft_show_elements
Signed direct upload — shell recipes, so a 200 MB reference streams from disk, not through the context window
Any public URL → reference
Interactive file picker
Small files without a shell
Verify an upload landed
Your media library
Reusable characters and styles
Jobs & catalogpopcraft_models_explore
popcraft_job_display
popcraft_show_generations
popcraft_projects
Capabilities, reference limits, aspect ratios, durations and credit costs per model
Result cards that refresh themselves — no polling loops
Revisit and chain past results
File results into projects
Skillspopcraft_get_workflow_instructions
popcraft_get_workflow_bundle_file
Production playbooks served live from the server — see below
Per-step reference modules
Accountpopcraft_balance
popcraft_show_plans_and_credits
popcraft_transactions
Credits and plan before spending
Plans and one-click top-up
Every charge and refund

Skills, served live

Ask for a deliverable rather than an asset and the agent loads the matching production recipe before it generates anything. Nothing to install, nothing to paste.

  • ugc-ad — one vertical 9:16 creator-style ad, up to 15 s, with speech and sound rendered natively in the clip.
    intake → creator identity → script → storyboard → realism pass → cost preview → one Seedance 2.0 clip → QA
  • thumbnail-pro — click-optimised YouTube / Instagram thumbnails with an identity lock for your own face.
    16 concept frameworks → one intake question → 11-block prompt → parallel variants → micro-edits → 4K upscale
  • h3-product-ad — minimalist, typography-led product ad on Hailuo 3, 5–15 s, with a native score.
    intake → copy → three anchor photos → storyboard → cost preview → one generation → QA
  • voiceover — narration that does not sound like TTS: voice lock, words-per-second budget, per-line direction.
    script rewrite → voice pick → line-by-line delivery tags
  • sd25-prompting / sd25-prompting-zh — ByteDance's official Seedance 2.5 prompt optimizer, served verbatim (English and Chinese editions); compiles your brief and references into one submission-ready prompt.
  • gpt-image-25-prompting / h3-prompting — model-specific prompt formats for GPT Image 2.5 (change-vs-preserve structure) and Hailuo 3 (structured format with sound and speech fields).

Cost preview on every spend

Every generate tool accepts get_cost: true and returns the exact credit price without submitting anything. Rates come from popcraft_models_explore; the numbers below are as of 2026-09-25 and apply before plan discounts.

Video modelCredits / second (720p)Image modelCredits / image
Seedance 2.0 Mini14Seedream 4.05
Seedance 2.0 Fast20GPT Image 2.5 Fast10
Seedance 2.026GPT Image 2.511
Seedance 2.5 · 480p draft42 · 24Nano Banana 215
Veo 3.1 Fast · Veo 3.115 · 58Nano Banana Pro · 4K20 · 35
Kling 3.0 Omni16
Wan 3.0 · Hailuo 320

Pro and Max include a daily Seedance 2.0 Fast & Mini allowance at no credit cost, and Nano Banana / Seedream images are unlimited from Plus upward — see unlimited-ai-video and pricing.


How a good agent run looks

  1. Pick the model — popcraft_models_explore: read its reference_limits, the video_input / image_input / audio_input envelope, durations and aspect ratios.
  2. Get references in — popcraft_media_import_url for anything with a public URL; otherwise popcraft_media_upload and run the returned shell recipe (curl -T file "<proxy_upload_url>"). Pre-check each file against the envelope — an out-of-envelope reference is rejected, or worse, wastes credits.
  3. Bind references in the prompt with the model's tokens — Image 1, Video 1, Audio 1 for the Seedance family; <Picture 1> for Hailuo 3.
  4. Preview, then spend — get_cost: true, show the number, then generate. The result card refreshes itself; do not poll in a loop.
  5. Chain — any result URL is a valid medias[].value for the next call: upscale it, animate it, edit it.

Recipes

Each recipe is a complete run — the tool calls, the prompts and the outputs, shown inline.

#RecipeWhat you get
01UGC ad from one product photoThe ugc-ad skill end to end, one real run: locked creator → script → storyboard → first frame → a 15 s Seedance 2.0 clip with native speech
02Thumbnail with identity lockthumbnail-pro with your own face held constant across variants
03Seedance 2.5 single takeTimestamped shot list, 480p draft, then popcraft_upgrade_video to 1080p
04Product image set with an exact wordmarkPackshot → lifestyle → background swap → poster → product loop, with the brand's exact wordmark
05Batch product imagesN products × K scenes in one loop, with consistency rules and cost math

FAQ

Does it need a paid plan? No. The connector works on the free tier (100 credits on sign-up). Some premium models are reserved for paid plans; the agent gets a clear message and can pick an alternative from the catalogue.
Where do generations go? Into your Popcraft account. Everything made over MCP appears in your web-app history for editing on the Canvas, upscaling or downloading; popcraft_show_generations lists them inside the chat.
How long does a job take? Images a minute or two (Midjourney longer); video a few minutes depending on model and length. Jobs are asynchronous — the agent gets a job id instantly and the card updates when the result lands.
Can I use my own images, video or audio as references? Yes — via popcraft_media_import_url for public URLs or the popcraft_media_upload shell recipe for local files. Generated results chain directly too.
What about real people in references? The Seedance 2.0 family runs a real-person verification on reference media before generating; a reference that fails (typically a recognisable public figure) is rejected before any charge.

popcraft.ai/mcp · Models · Pricing · Unlimited
MIT licence — see LICENSE. Model names belong to their owners.

Source: README.md at commit 1a0a5d0

Tools

0
This server asks you to sign in, so SourceWeft could not read its tools without an account and has not indexed them. What a connected account sees can depend on that account and its permissions.

Version history

2
  1. v1.0.1LatestSep 29, 2026
  2. v1.0.0Sep 28, 2026

Score

85% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 daysRequired25 pts
  • Auth declaredRequired10 pts
  • Tools listed15 pts

    Answer tools/list without a credential, or ship a package that starts in a sandbox without one.

  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts