Skip to main content
Glama
README.md
# AutoShort MCP

[![Official MCP Registry](https://img.shields.io/badge/MCP%20Registry-org.autoshort%2Fautoshort-blue)](https://registry.modelcontextprotocol.io/v0.1/servers?search=autoshort)

Turn an idea into a faceless short video: script, voiceover, captions and scenes, from any AI assistant that speaks MCP. The primary way to use it is the **hosted remote MCP server** of [AutoShort](https://autoshort.org) (nothing to install). This repository also holds the MIT-licensed source of the **local stdio server** (`mcp-server-autoshorts` on npm), a thin proxy that forwards `tools/list` and `tools/call` to the hosted endpoint with your API key.

- **Endpoint:** `https://autoshort.org/api/mcp` (Streamable HTTP)
- **Registry name:** `org.autoshort/autoshort` ([Official MCP Registry](https://registry.modelcontextprotocol.io/v0.1/servers?search=autoshort))
- **Auth:** OAuth 2.0 with PKCE and dynamic client registration (discovered from `/.well-known/oauth-protected-resource`), or an AutoShort API key as `Authorization: Bearer sk_live_...`
- **Docs and API keys:** https://autoshort.org/en/developers
- **Also on npm** as a stdio proxy: `npx -y mcp-server-autoshorts` with `AUTOSHORT_API_KEY` set (source in `src/`)

## What you need

An AutoShort account. The server is a connector to your own account, not a free video generator: drafts are free, renders spend credits from your balance. AutoShort's current plans: a 3-day trial for $2.99 (non-refundable, includes 150 credits), then from $19/month; credit packs start at $2.99. See https://autoshort.org/en/pricing for the live prices.

## Connect

**Claude Code**

```bash
claude mcp add --transport http autoshort https://autoshort.org/api/mcp
```

then run `/mcp` and authenticate `autoshort` (a browser opens: sign in and allow).

**Gemini CLI**

```bash
gemini extensions install https://github.com/airdropfi/autoshort-gemini-extension
```

**VS Code** (`.vscode/mcp.json`)

```json
{
  "servers": {
    "autoshort": { "type": "http", "url": "https://autoshort.org/api/mcp" }
  }
}
```

**Any other client that supports remote (Streamable HTTP) servers**: add the URL above. If the client cannot do OAuth, create an API key in the AutoShort dashboard and send it as `Authorization: Bearer sk_live_...`.

## What the tools cover

54 tools, with read-only / destructive annotations on each:

| Area | Examples |
|---|---|
| Draft and render | draft a script, review and edit it, revise a scene, price it, render it, cancel, track status |
| Formats | faceless videos, chat-style videos, product-page ads, image-to-video, motion clips, UGC-style campaigns |
| Series | plan a multi-episode character series, character sheets, per-episode quotes, pause/resume |
| Account | credit balance, usage, limits, cost estimate, subscription, webhooks, API keys |
| Publishing | list connected social accounts and publish a finished video |

Rules the server enforces: a draft is free and renders nothing; a render is priced first (`autoshort_estimate_cost` / `autoshort_quote_video`) and spends credits only when you confirm it; real or famous people, protected characters and sexual content are refused. Videos use AI-generated images and voice: switch on the platform's AI-content label when you post. No result or performance is promised.

## Related

- Claude Code plugin: https://github.com/airdropfi/autoshort-claude-plugin
- Gemini CLI extension: https://github.com/airdropfi/autoshort-gemini-extension
- Privacy: https://autoshort.org/en/privacy · Terms: https://autoshort.org/en/terms · Support: support@autoshort.org

## Run the stdio server from source

```bash
npm ci && npm run build
AUTOSHORT_API_KEY=sk_live_... node dist/index.js      # or: docker build -t autoshort-mcp . && docker run -i -e AUTOSHORT_API_KEY autoshort-mcp
```

Without an API key the server still starts and answers `tools/list` from an offline snapshot of the hosted server's public tool list (`src/tools-snapshot.ts`; no network, no account), but every tool call is refused until a key is set. With a key, the tool list is fetched live from the hosted server, so it is always current.

## About this repository

It holds the registry metadata (`server.json`, identical to the published registry entry), documentation and the source of the stdio proxy (`src/`, `tests/`, `Dockerfile`). The hosted server's own code (the tool registry, billing and rendering) is closed source and runs at autoshort.org. MIT license covers the contents of this repository.

TDQS

A3.8/5.0

Scored across 54 tools

Disambiguation3/5

The set has six or seven overlapping video-creation entry points (generate_video, generate_chat_video, generate_product_ad, generate_image_to_video, generate_motion_clip, draft_video, character/UGC episode tools), which forces an agent to rely on lengthy embedded ROUTING notes rather than clear tool boundaries. The descriptions do actively disambiguate (drafts vs rendered, stills vs motion clips, narrated vs silent), so misselection is avoidable but expensive with 54 tools. Pricing is also split across estimate_cost, quote_video and quote_character_episode in ways an agent must learn.

Naming Consistency4/5

Every tool uses the autoshort_ prefix with snake_case verb_noun (get_, list_, create_, update_, render_, publish_, quote_), which is highly predictable. Minor strain comes from nested namespaces (ugc_*) and near-synonym verbs: estimate_cost vs quote_video, buy_credits vs get_buy_credits_url vs setup_auto_recharge, plan_drama_season vs save_season_plan.

Tool Count2/5

54 tools is far beyond a well-scoped surface, even for a platform spanning generation, drafts, publishing, billing, webhooks, series, character sheets and season plans. Several clusters (six generation entry points, four credit/purchase tools, many UGC sub-steps) could be collapsed into parameterized tools, and the volume itself becomes a discovery burden.

Completeness4/5

Coverage is unusually thorough: drafting, editing, rendering, cancellation, publishing per platform, credits/billing, spending limits, webhooks, series autopilot, character sheets, season plans and performance stats. Gaps are modest — no video deletion or metadata update tool, and no delete/archive series operation (only pause), which agents can mostly work around.

Maintenance

ActivityMaintained
ResponsivenessOutdated data