Connect AI Assistants

OyeFilmy MCP Server

OyeFilmy MCP is a remote server built on the Model Context Protocol. Once connected, your AI assistant (Claude, ChatGPT, Cursor, Windsurf, Claude Code) can generate images and videos, synthesize audio, build 3D models, upscale assets, and search your creation history directly from your chat prompt.

Remote Server Endpoint: https://api.oyefilmy.com/api/mcp

Uses Streamable HTTP (SSE / JSON-RPC 2.0). No local bridge or binary installation required.

Getting started

person

1. OyeFilmy Account

Any OyeFilmy account works. MCP tools share your existing credit balance.

power

2. Add Server

Add https://api.oyefilmy.com/api/mcp in your AI client's MCP settings, with an Authorization header.

verified_user

3. Add Your API Key

Send an of_live_… key as a bearer token. Create one on the API Keys page.

auto_awesome

4. Start Creating

Ask your assistant to generate, upscale, or search. It picks the right tool automatically.

Connect your client

Claude Web — Connectors

Step 1 of 6

Claude Chat Interface

Step 1: Open Customize Menu

Click your profile avatar in the bottom-left corner and select Customize.

Open Claude and click your profile avatar in the bottom-left corner, then select Customize.
Go to Connectors and click Add connector.
Select Add custom connector.
Enter Name: OyeFilmy and URL: https://api.oyefilmy.com/api/mcp.
This is where it stops today.The connector will save and list the tools, but Claude Web authenticates custom connectors over OAuth and offers no field for a bearer token — so there is no way to attach your key.
Once OAuth ships you will approve access here and can ask Claude: "Generate a futuristic drone video in 4K using Seedance 2.0".

Claude Web is not supported yet. OyeFilmy authenticates MCP callers with an of_live_…bearer key, and Claude Web's connector UI has nowhere to put one. Use Claude Code, Cursor, Windsurf or ChatGPTtoday — all four accept a header. OAuth support is planned, and this page will change when it lands.

Available MCP tools

The MCP server advertises these tools via tools/list, which is generated from the live model routing maps and is the authoritative list — this table is a summary. Your agent calls them autonomously based on your prompts.

Tool Name	Category	Description
`account_balance`	Account	Current credit balance and plan tier
`cost_estimate`	Account	Price a generation without running it — the exact credits it would be billed, and whether the key’s spend cap covers it
`images_generate`	Image Generation	Generate images from a prompt, with aspectRatio, resolution, imageQuality, styleRef, charRef and brandId
`video_generate`	Video Generation	Generate video (T2V, I2V, start-end frame) with prompt, aspectRatio, resolution and duration
`audio_generate`	Audio & Speech	Generate music (MiniMax Music, Seed-Audio), speech (ElevenLabs, xAI TTS, Gemini TTS) or lyrics
`models3d_generate`	3D Generation	Generate a textured GLB mesh from text or an image, via Hunyuan3D or Seed3D
`images_upscale`	Image Editing	AI image upscaling — 2x, 3x or 4x, returned as PNG
`video_upscale`	Video Editing	AI video upscaling to 1080p or 2K
`images_remove_background`	Image Editing	Extract main subject with transparent PNG alpha cutout
`creations_search`	History & Search	Search user's historical generation library by prompt keyword, type, and date
`creations_get`	History & Search	Retrieve detailed metadata, model settings, and download URL for an asset ID
`creation_status`	Jobs & Polling	Poll an in-flight generation. Holds at "processing" until the output URL exists, so a completed result always carries one
`creations_wait`	Jobs & Polling	Block until a generation finishes. Good for images; for video prefer creation_status
`models_list`	Catalog	The live model catalogue: counts per modality and the keys worth reaching for first
`brand_dna_list`	Brand DNA	List active Brand DNA profiles for brand-consistent AI generation
`film_vocabulary`	Direction	The moods, ad formats, narrative structures and shot archetypes the film tools accept. Free
`style_bible_compile`	Direction	Compile a mood into the craft grammar header every shot must carry — byte-identical, budget-fitted, brand conflicts surfaced. Free
`film_plan`	Direction	Turn a brief or screenplay into a shot plan whose durations sum to exactly the target length. Free
`validate_prompt`	Direction	Check a prompt fits the model ceiling and names no director or cinematographer, before you pay to render it. Free

Directing a sequence, not just calling a model

Any assistant can call an image or video model. What it cannot usually do is make eight separately-rendered shots look like one film. Four tools exist for that, they cost no credits, and they are the reason to point your agent here rather than at a raw model API.

1. film_vocabulary        → pick a moodId (never a director's name)
2. style_bible_compile   → one craft header, byte-identical for every shot
3. film_plan             → shot list whose durations sum to exactly your total
4. validate_prompt       → catch a ceiling overrun or a stray name before paying
5. video_generate ×N     → render each shot, header first, previous last frame as inputUrl

Why the header matters. It is deterministic: the same mood and target model produce the same string, character for character. Your agent can compile it once, keep it in its own context, and render shot 7 an hour after shot 1 — and the two still cut together. Two shots of one project carrying different headers is the defect this removes.

Why the durations matter. film_plan absorbs its own rounding into the peak shot, so the shot lengths sum to the length you asked for. Render each shot at its own seconds and the sequence is on time. The response says so explicitly in durationContract rather than leaving your agent to add it up.

Director names are not an input. Pass a moodId. The compiler translates a craft reference into concrete lens, lighting, palette, movement and sound language, and no director or cinematographer proper noun is allowed into an emitted prompt — validate_prompt refuses one. Naming the person is both a rights risk and, in practice, a vaguer prompt than the grammar it stands for.

Frequently asked questions

How does OyeFilmy connect to AI agents?

OyeFilmy runs a remote server that speaks the Model Context Protocol over Streamable HTTP. Any MCP-compatible client connects directly to https://api.oyefilmy.com/api/mcp and calls OyeFilmy tools right from the chat.

Do I need an API key?

Yes. Authenticate with an OyeFilmy API key (of_live_…) sent as a bearer token — create one on the API Keys page. initialize and tools/list answer without credentials so a client can discover the tools, but every tool that touches your account or spends credits requires the key, and it must carry the right scope for what you ask (generate:image, generate:video and so on). OAuth is not supported yet.

How does credit billing work?

MCP tool invocations draw from the same credit balance as your main OyeFilmy account, priced by model tier, resolution and — for video — duration. A failed generation is refunded automatically. Your assistant can call cost_estimate to get the exact figure before it spends anything, so you can approve a render rather than discover it.

How do I stop an agent from spending everything?

Give the key a monthly spend cap when you create it on the API Keyspage. A request that would cross the cap is refused before any credits move, and the rest of your balance stays out of reach — so a key you hand to something running unattended is limited to what you decided, not to what your account holds. Caps reset on the 1st, and the key's scopes independently limit which kinds of model it can run at all.

Can my agent search my past creations?

Yes! The creations_search and creations_get tools allow your agent to look through your historical generations and use previous assets as inputs for video, upscaling, or editing.