Composer rounds for agents
Ten MCP tools for Composer. An agent can now run a brand’s daily rounds the way the app deals them:
todays_round, reveal_round, shuffle_round, approve_play, undo_approve, pass_play, plus the Library shelf with save_to_library, list_library, remove_from_library and download_library. Round actions are logged as an agent’s, not a person’s.The MCP server now exposes 55 tools. See the Percify MCP server.Batches, personas and spend alerts
Server-driven batches.
POST /v1/batches runs one model over up to 50 prompts, and because the server drives it you can start a batch and hang up. budgetCredits is a hard ceiling checked before each submission, and polling a batch that lost its driver resumes the prompts it had not reached. See the batches API.Avatar personas hold a character, and any model can use one. Persona images can be screened as a batch before any of them are written.A choice of lip-sync model, and spend alerts on API keys.Composer deals rounds instead of a wall
Plays are now dealt in rounds of three, face down. Reveal turns the three over, Approve locks a card, and Shuffle rerolls only the cards still open. Any single approved card can be scheduled on its own.This replaced the sixteen-card wall. Measured over the ten days before the change, eleven brand owners reached the wall, eight hovered, and three ever kept a card: sixteen at once reads as a catalogue to browse rather than a decision to make. See Composer.
What the API needs to be true
A webhook signing secret you can read from the API, a complete OpenAPI document, and full tool definitions for MCP clients. See Webhooks and errors and limits.
Brand OS for agents
Brand OS over MCP: read a brand from its website, write its words into a proven short video format, render it, and publish to connected social accounts, with an autopilot that learns from each post. Agents can also wait on any job server-side and list past generations.
MCP gets its own origin
The MCP server moved to mcp.percify.io, served by the same API. Connections set up with the earlier URL keep working.
Composer for agents and scripts
The Composer API at
api.percify.io/v3/composer/api, on the same pk_live key and the same Scale-and-above plan gate as the rest of the API. See the Composer API.GPT Image 2.5
GPT Image 2.5 Flare and Sunburst at 2 credits a run, with Flare Edit and Sunburst Edit at 4. See every model and its price.
Realtime video and Gemini TTS
Realtime Video from text or from an image at 60 credits, and Gemini TTS at 4 credits. See Realtime.
Seedance 2.5 and Wan 2.7
Seedance 2.5 text-to-video and image-to-video at 120 credits, the image-to-video Turbo at 150, Wan 2.7 Video Edit at 140, and P-Video Replace at 26.
How to follow changes
- Model prices and clip lengths change as models are added.
GET /v1/modelsalways returns the current catalogue, and the Playground shows each model’s price before you run it. - Every docs page carries the date it was last updated, under its title.
- Breaking API changes are avoided rather than versioned away: routes are added, and existing ones keep answering. Where a URL had to move, a redirect keeps the old one working.
Related
Every model and its price
The current catalogue, grouped by what it makes.
The Percify API
Base URL, keys and plans.
The MCP server
All 55 tools, grouped by job.
Pricing
Plans, credit packs and billing.