Add the Percify MCP server to Claude, ChatGPT, Cursor or any MCP-compatible client. 25+ models for image, video and voice — plus shot-level video intelligence and AI replication. Sign in with your Percify account, no API key required.
Step 1 — Copy the server URL
https://api.percify.io/v3/mcpYou'll paste it in the next step.
Step 2 — Add the connector
In Claude → Settings → Connectors, choose Add custom connector, name it Percify and paste the URL.
Step 3 — Connect & sign in
Click Connect — a Percify sign-in window opens. Log in and hit Authorize. That's it: ask Claude to generate something.
claude mcp add percify --transport http https://api.percify.io/v3/mcp --header "Authorization: Bearer pk_live_KEY"{ "mcpServers": { "percify": {
"url": "https://api.percify.io/v3/mcp",
"headers": { "Authorization": "Bearer pk_live_KEY" } } } }API keys live at app.percify.io → Developer · REST reference at docs.percify.io
Six workhorses from a live catalog of 25+ — list_models always has the current roster, schemas and pricing.
ByteDance
The workhorse for faithful image-to-video — conditions every clip on a source frame.
OpenAI
Cinematic image-to-video with standout physical realism, up to 12s per clip.
Kling AI
Drives your character with a reference video’s real movement — the Motion Match engine.
OpenAI
Instruction-following image generation and editing with best-in-class text rendering.
Edit and blend photos conversationally — identity-preserving product and face work.
Zyphra
Voice cloning from a short sample, speaking any script in 20+ languages.
Video Analyzer
Feed analyze_video a TikTok, YouTube Shorts or Reels URL and get a shot-level blueprint back as structured JSON: per-clip generation prompts, per-second action beats, camera framing and movement, on-screen text with timestamps, and the VO transcript.
“Analyze this short and tell me why the hook works” → hook / build / payoff structure with exact timings.
Video Replicator
replicate_video renders each blueprint clip through your chosen model, stitches them, optionally burns the original’s captions, and returns a branded side-by-side comparison plus a public share link. Pick any subset of clips or preview just the hook.
“Replicate this ad but make the athlete a golden retriever” → full replica with a compare video.
Star In It
Pass a characterImage and it stars in every clip — as the conditioning frame on image-to-video models or the character for Kling Motion Match, which drives your subject with the original’s real movement.
Your product photo + a trending unboxing short → the same video, selling your product.
Generate Anything
generate runs any catalog model — list_models always returns the live roster with schemas and credit pricing, and get_model gives a worked example for each.
“Make me a 9:16 clip of a neon jellyfish city” → generation id → poll → mp4 URL.
Talk & Voice
InfiniteTalk animates a portrait to any audio from 2 credits per second; Zonos 2 clones a voice from a sample and speaks any script in its tone — the audio layer for UGC ads, dubs and talking avatars.
A selfie + a script → a talking-head video with a cloned voice.
Agent Commerce
check_usage returns the live remaining balance, monthly spend and a top-up link. Failed generations refund automatically, retried requests honor Idempotency-Key so nothing double-charges, and completion webhooks are signed with X-Percify-Signature.
“How many credits do I have?” → an exact number, not a guess.
Percify exposes a standard MCP server (Streamable HTTP) at https://api.percify.io/v3/mcp. Any client that speaks the Model Context Protocol connects with that one URL — OAuth 2.0 discovery, dynamic client registration and PKCE are handled automatically.
Claude (web, desktop and mobile) connects via custom connectors with OAuth. Claude Code, Cursor, Codex and any client that supports custom headers can authenticate with a pk_live API key instead. If it speaks MCP, it works with Percify.
No — in Claude, click Connect and sign in with your Percify account. API keys are only needed for clients without a browser sign-in flow, and can be created in the Percify developer console.
The catalog is live and always current via list_models. Today it spans 25+ models: Seedance 2.0, Sora 2 and Sora 2 Pro, Kling Motion Control, GPT Image 2, Flux 2, Nano Banana 2, SeeDream 5, Grok Imagine, Wan 2.2 Animate, InfiniteTalk lip-sync, Zonos 2 voice cloning and more.
Images up to 1080p+, video clips up to 15 seconds per generation (stitchable into full shorts), lip-synced talking avatars, cloned voices and multilingual speech — plus complete AI replicas of existing short-form videos with side-by-side comparisons.
Everything is credit-based and charged to the connected Percify account at the same prices as the app. Cheap image models start at 2 credits, video from about 10 credits per clip. Failed generations refund automatically, and check_usage returns your balance plus a top-up link.
Images typically finish in seconds; video clips take one to a few minutes depending on the model. Generation tools return an id immediately — agents poll for the result or pass a webhook that fires once on completion, signed with X-Percify-Signature.
Percify’s flagship: analyze_video turns any short into a shot-level creative blueprint, and replicate_video regenerates it clip by clip — optionally starring your own character or product — then stitches the result and renders a side-by-side comparison with the original. It is an ad-remix engine exposed as three tool calls.
Yes — OAuth connections can be removed from your Percify account at any time, and API keys can be revoked in the developer console. Each connection only ever spends the credits of the account that approved it.