# ViewPress

> ViewPress makes short-form videos with AI: generate images, video clips,
> voiceover, music, and SFX; compose them on a timeline; caption, enhance, and
> publish to TikTok, Instagram, and YouTube. The whole surface is available to
> agents over MCP and a REST API.

## Connect an agent (MCP)

MCP server: https://viewpress.ai/api/mcp (streamable HTTP).

Claude, Claude Code, Cursor, and VS Code sign in with OAuth — no API key
needed. For Claude Code:

    claude mcp add --transport http viewpress https://viewpress.ai/api/mcp

then run /mcp, select viewpress, and choose Authenticate. Header-configurable
clients can send `Authorization: Bearer vp_...` instead (create keys at
https://viewpress.ai/api-keys).

## API

- [API reference](https://viewpress.ai/docs/api): every endpoint, auth,
  credits, errors, and the timeline (tracks) model
- [OpenAPI spec](https://viewpress.ai/api/v1/openapi.json): machine-readable,
  generated live from the server's own validation schemas
- [Render body schema](https://viewpress.ai/api/v1/renders): full JSON schema
  of the timeline composition payload (GET, no auth)

## How it works

- All generation is async: a POST returns `{ assetId, status: "pending" }` —
  poll `GET /api/v1/assets/{assetId}` until `ready`.
- Generation debits credits, auto-refunded on failure. Rate limit: 120
  requests/minute per account, one window shared across REST, MCP, API keys,
  and OAuth connections. Over it returns 429 with a `retry-after` header
  (seconds).
- Every generated asset attaches to a video's chat in the web app, so a person
  can open the video and continue the work where the agent left off.