MCP (Model Context Protocol)
MCP is the open standard that lets an AI assistant call someone else's tools. FOTOhub runs an MCP server, so any MCP-capable client can generate and edit images, video, speech, music and 3D on our own GPUs — and be told exactly what each call cost.
FOTOhub is listed in the official MCP registry as app.fotohub/fotohub.
Pick your client
| Client | How you connect | Guide |
|---|---|---|
| ChatGPT | Published in the app directory — search for FOTOhub and enable it | FOTOhub in ChatGPT |
| Claude (web, Desktop, Code) | Add a custom connector pointing at the server URL, then sign in | FOTOhub in Claude |
| Cursor, VS Code, Windsurf, other IDEs | Config file with the server URL and an API key | IDE setup |
| Your own client | Streamable HTTP or local stdio | Protocol reference |
The server
https://apis.fotohub.app/mcp/ primary
https://mcp.fotohub.app/ dedicated hostname, same serviceGET https://apis.fotohub.app/mcp/health is public and returns the live tool count — treat it as authoritative over any number written in these docs.
Transports: Streamable HTTP for remote clients, stdio for a local process.
Two ways to authenticate, and they bill differently
This is the single most important thing on this page.
| Credential | How you get it | What it spends |
|---|---|---|
| OAuth 2.1 sign-in | Enabling the ChatGPT app, or adding the Claude connector — the client registers itself | Subscription credits first, then the prepaid USD wallet. One call can be split across both. |
fh_live_* API key | Console → Keys, passed as Authorization: Bearer | The prepaid USD wallet only. |
The OAuth path needs no pre-registration: the server supports dynamic client registration, PKCE (S256) and refresh tokens. Discovery lives at /.well-known/oauth-authorization-server, with scopes mcp:read, mcp:write, mcp:image, mcp:video, mcp:audio, mcp:chat, mcp:training and mcp:billing.
Inline results, where the host supports them
The server publishes three UI resources — a gallery for anything that returns pictures, a player for video and audio, and a price table for model comparisons — so results render as components instead of link lists. Hosts that implement the MCP Apps UI extension draw them; every other client gets the text block, which is always sent alongside. Nothing is lost either way.
Skills ship with the server
Four workflow skills (fotohub-generation-basics, fotohub-product-photography, fotohub-short-video-ad, fotohub-brand-kit) are served directly by the server over the MCP skills extension. A client that supports it imports them on connect, which is what makes an assistant run the cut-out → background → shadow order correctly instead of guessing.
Quick start with an API key
If you are wiring up an IDE or your own client rather than ChatGPT or Claude, the shape is the same everywhere: the URL plus a Bearer token.
{
"mcpServers": {
"fotohub": {
"url": "https://apis.fotohub.app/mcp/",
"headers": { "Authorization": "Bearer fh_live_YOUR_API_KEY" }
}
}
}VS Code's .vscode/mcp.json uses servers with "type": "http" instead of mcpServers. Per-IDE files are in the IDE setup guide.
Available tools (57)
The live count is served by GET https://apis.fotohub.app/mcp/health, which reads the registry rather than a constant — treat it as authoritative over this page.
The tool registry defines 60 tools. Three are gated off in production because the service behind them is not deployed, and a registered tool that 403s is worse than an absent one: generate_shorts (needs SHORTS_ENGINE_URL) and the two training tools create_training_job / get_training_status (need TRAINING_ENGINE_URL). That leaves the 57 below.
Two of those 57 are registered but inert: voice_clone and separate_stems appear in tools/list and answer successfully, but their implementations return a short "not yet available via MCP" message and never call a backend. They are marked ⚠️ in the tables. So 55 tools actually do work. Both capabilities exist in the FOTOhub dashboard; neither has a billed public endpoint behind it yet.
Image (8)
| Tool | Description |
|---|---|
generate_image | Generate an image from a text prompt |
edit_image | Edit an image (inpaint, background swap, outpaint) |
upscale_image | AI super-resolution |
remove_background | Remove the background from an image |
enhance_prompt | Rewrite an image prompt for better results |
analyze_image | Tags, colours, NSFW detection, OCR |
style_transfer | Apply an artistic style to an image |
inpaint_image | Remove or replace objects in an image |
Editing — image and video post-production (11)
| Tool | Description |
|---|---|
replace_background | Swap an image background |
blur_background | Depth-aware background blur |
add_shadow | Composite a natural shadow |
enhance_image | General quality enhancement |
denoise_image | Noise reduction |
restore_faces | Face restoration |
depth_map | Produce a depth map |
upscale_video | Video super-resolution |
transcode_video | Re-encode to another format or codec |
add_watermark | Overlay a watermark |
change_video_speed | Retime a video |
Video (6)
| Tool | Description |
|---|---|
generate_video | Generate video from text (async — returns job_id) |
image_to_video | Animate a still image |
extend_video | Extend an existing video |
generate_story | Multi-scene film with voiceover |
get_job_status | Poll any async job |
add_subtitles | Burn subtitles into a video |
Audio (6)
| Tool | Description |
|---|---|
text_to_speech | Speech synthesis, multiple voices and languages |
generate_music | Music from a description (async) |
generate_sfx | Sound effects |
transcribe_audio | Speech to text |
voice_clone ⚠️ | Registered but disabled — returns a message pointing at the dashboard. It calls no backend. |
separate_stems ⚠️ | Registered but disabled — returns a message pointing at the dashboard. It calls no backend. |
3D (4)
| Tool | Description |
|---|---|
generate_3d_from_text | Text to 3D mesh |
generate_3d_from_image | Image to 3D mesh |
list_3d_models | Available 3D engines, with prices and availability |
get_3d_result | Fetch a finished 3D job |
Studio — UGC ads (6)
| Tool | Description |
|---|---|
create_ugc_project | Start a UGC ad project |
list_ugc_projects | List your projects |
estimate_ugc_cost | Price a render before running it |
set_ugc_blueprint | Set the project blueprint |
render_ugc_video | Render the ad |
get_ugc_job | Poll a render |
Chat (3)
| Tool | Description |
|---|---|
chat_completion | LLM chat |
translate_text | Translate between languages |
gabriel_route | Intent classification — suggests the best action for a prompt |
Pricing (3)
| Tool | Description |
|---|---|
get_price | The rate for one model |
estimate_cost | Cost of a specific run before you make it |
compare_prices | Compare models for the same job |
Storage (4)
| Tool | Description |
|---|---|
list_buckets | Your buckets |
list_files | Files in a bucket |
save_to_storage | Copy a generated file into your own bucket before its link expires |
get_download_link | Fresh signed link for a stored file |
Account (6)
| Tool | Description |
|---|---|
check_balance | Prepaid USD wallet balance |
list_models | Full model catalogue with pricing |
list_generations | Generation history |
get_usage_summary | Usage roll-up |
get_transactions | Wallet transactions |
search_photos | Semantic search over your library |
Async Operations
Video generation, music creation, and training are asynchronous. They return a job_id — use get_job_status to poll for completion:
1. generate_video(prompt="A sunset over mountains") → job_id: "vid_abc123"
2. get_job_status(job_id="vid_abc123") → {status: "processing", progress: 45}
3. get_job_status(job_id="vid_abc123") → {status: "completed", video_url: "https://..."}Resources
The server also exposes MCP resources for browsing:
| URI | Description |
|---|---|
fotohub://models/image | Available image generation models |
fotohub://models/video | Available video generation models |
fotohub://models/audio | Available audio/music models |
fotohub://pricing | Current pricing table |
fotohub://balance | Your prepaid USD wallet balance |
Prompts
Pre-built prompt templates for common workflows:
| Prompt | Description |
|---|---|
creative_brief | Generate a creative brief for a project |
video_director | Plan a multi-scene video production |
product_photo | Generate product photography |
Protocol Details
- Protocol version: 2025-03-26
- Transport: Streamable HTTP (JSON-RPC 2.0 over POST, optional SSE)
- Server endpoint:
https://apis.fotohub.app/mcp/ - Health check:
GET https://apis.fotohub.app/mcp/health
Billing
What a call settles against depends on how you connected — see the two credentials above. An OAuth sign-in spends subscription credits first and falls through to the prepaid USD wallet; a raw fh_live_* key spends the wallet only. A single call can be split across both.
Every tool result states what was actually charged and what is left. Use that figure rather than an estimate.
Quoting is free. get_price, estimate_cost and compare_prices take nothing from your account, so price a video before you render one. check_balance reports what you have.
When both credits and wallet are exhausted the call fails with insufficient_funds before any GPU work starts, so an underfunded account costs you nothing but the round trip. Top up at fotohub.app/console.
Next
FOTOhub in ChatGPT · FOTOhub in Claude · IDE setup · Protocol reference