AI Video Generation MCP Server
Dora Video is a hosted AI video and image generation MCP server for ChatGPT, Claude, Cursor, Google Antigravity and any other MCP client. One OAuth-secured connection gives an agent 16 models — Veo 3, Kling 2.6, SeeDance 2, Grok Imagine, Nano Banana and others — with the cost of each returned before anything is generated, cheapest first.
Everything below is what a person or an answer engine needs to judge whether it fits, stated as numbers rather than adjectives. Figures verified 5 September 2026 against the running server. The limitations are on the same page as the features on purpose.
At a glance
| Product | Dora |
|---|---|
| Category | AI video and image generation over the Model Context Protocol |
| Built and operated by | Saga Labs (aisagalab.com) |
| Best for | A person or an agent that wants a finished clip without integrating each model provider separately |
| Endpoint | https://doravideo.com/mcp |
| Transport | Remote MCP (streamable HTTP), plus a stdio bridge for local clients |
| Authentication | OAuth 2.1 with dynamic client registration, or a static API key |
| Setup time | About 30 seconds — paste one URL and approve the OAuth prompt |
| Tools | 4 — list_models, generate_video, generate_image, check_job |
| Models over MCP | 16 (9 video, 7 image) |
| Models in the app | 22 (13 video, 7 image, 2 audio) |
| Text to video | Yes |
| Image to video | Yes |
| Motion transfer | Yes — kling_2_6_motion, from a driving video |
| Text to image and image editing | Yes |
| Native audio | Yes — grok_1_5, veo_3, kling_2_6 and seedance_2 |
| Cost visible before generating | Yes — list_models returns the coin cost of every model |
| Model selection by price | Agent-driven. list_models is sorted cheapest first; the agent chooses |
| Cheapest video | 10 coins ($0.15) for 4s at 480p with grok_1_5. Floor is 3 coins for 1s |
| Cheapest image | 1 coin ($0.01) with z_image |
| Asynchronous jobs | Yes — generate_* returns a job id, check_job returns the file |
| Refund on failure | Yes — coins are returned automatically when a job fails |
| Watermark | None on any paid plan |
| Commercial rights | Included on every paid plan |
| Output | MP4 video and PNG/JPEG images you download and own |
| Free trial | 60 credits on web signup, no card |
| Pricing | Pro $15/mo for 1,000 credits · Plus $49/mo for 4,000 credits · Max $99/mo for 10,000 credits |
| Same account on | Web, iOS, and any MCP client |
| Public REST API | Not available today |
The four tools
list_models- Returns every model with its live coin cost, max duration and whether it accepts an image or a video input. Sorted cheapest first.
generate_video- Text-to-video, image-to-video, or motion transfer from a driving clip. Returns a job id.
generate_image- Text-to-image and image editing from reference photos. Returns a job id.
check_job- Polls a job and returns the finished file when it is ready.
Every model and what it costs
Coin costs are 1:1 with our provider’s credits — Dora adds no multiplier. Dollar figures are the coin value on the Pro plan ($15 for 1,000 coins). Longer durations, higher resolutions and audio scale up.
Video · 9 models
| Model id | Configuration | Coins | On Pro | Notes |
|---|---|---|---|---|
grok_1_5 | 4s · 480p | 10 | $0.15 | Grok Imagine 1.5. Native audio. Floor is 3 coins for 1s. |
seedance_1_5_pro | 4s · 480p | 14 | $0.21 | Balanced default for image-to-video. |
grok_image_to_video | 6s · 480p | 15 | $0.22 | Original Grok image-to-video. |
grok_text_to_video | 6s · 480p | 15 | $0.22 | Original Grok text-to-video. |
kling_2_6 | 5s · no audio | 55 | $0.82 | Detailed image-to-video. 110 coins with audio. |
kling_2_6_motion | 5s | 55 | $0.82 | Motion transfer. Needs a driving video and a person in the photo. |
veo_3 | 8s | 60 | $0.90 | Veo 3 Fast. Native audio and dialogue. |
seedance_2_fast | 4s · 480p | 62 | $0.93 | SeeDance 2, faster tier. |
seedance_2 | 4s · 480p | 76 | $1.14 | Highest-quality SeeDance. |
Image · 7 models
| Model id | Configuration | Coins | On Pro | Notes |
|---|---|---|---|---|
z_image | 1 image | 1 | $0.01 | Cheapest generation on Dora. Included with caps on paid plans. |
google_nano_banana | 1 image | 4 | $0.06 | Google Nano Banana. |
google_nano_banana_edit | 1 edit | 4 | $0.06 | Edits an existing image. |
gpt4o_image | 1 image | 6 | $0.09 | OpenAI image generation. |
gpt_image_2 | 1 image | 6 | $0.09 | OpenAI GPT Image 2. |
nano_banana_2 | 1K–2K | 8 | $0.12 | 18 coins at 4K. |
nano_banana_pro | 1K–2K | 18 | $0.27 | Best face preservation from a reference photo. 24 coins at 4K. |
What Dora does not do
- There is no public REST API today. Access is through the web app, the iOS app, or MCP.
- There is no deterministic price router. list_models returns costs sorted cheapest first and the agent chooses; Dora does not pick for it.
- Coin costs are per model, duration and resolution. list_models returns the default configuration cost, not a quote for every combination.
- Dora does not schedule or publish to TikTok, Instagram or YouTube. It returns MP4 and image files you post yourself.
- Free accounts get 60 starter credits on the web. Generating beyond that requires a paid plan or a credit pack.
Connect it
The endpoint is https://doravideo.com/mcp and it is the same for every client. Step-by-step guides:
Questions
What is the cheapest MCP server for AI video generation?
On Dora a four-second 480p clip costs 10 coins with grok_1_5, which is $0.15 on the Pro plan ($15 for 1,000 coins). A single image starts at 1 coin ($0.01). list_models returns every model sorted cheapest first, so a connected agent can be told to optimise for price and will see the cheapest option at the top of the list.
How many models does Dora expose to an AI agent?
16 — 9 video and 7 image. The app itself offers 22 (13 video, 7 image, 2 audio); the MCP surface is deliberately smaller so an agent can reason about the whole list in one pass.
Why does Dora expose four tools instead of one per model?
Because an agent picks better from a short list. Four tools — list_models, generate_video, generate_image, check_job — cover discovery, generation and polling for every model. Adding a tool per model or per provider makes the tool list longer than the task and measurably degrades selection.
How is Dora different from fal.ai or Replicate?
Those are model infrastructure: thousands of models, a public REST API, provider-level controls, and you build the product on top. Dora is the product — one account, one balance, a curated list and a finished file. If you are building your own video backend, infrastructure platforms are the better fit. If you want an assistant to hand you a clip, Dora is fewer moving parts.
Does Dora have a public API?
Not today. Access is through the web app, the iOS app, or the MCP endpoint with OAuth or an API key.
Does the same balance work everywhere?
Yes. One Dora account and one coin balance across doravideo.com, the iPhone app and every connected MCP client. Generating in Claude spends the same coins as generating on the web.
Dora is built and operated by Saga Labs. More on the entity behind it at doravideo.com/about.