MCP API Integration Guide

What is MCP?

MCP (Model Context Protocol) is a standard protocol that allows AI agents (Claude, GPT, Cursor, etc.) to directly call tools from external services. By connecting to the Autoagens MCP server, you can create videos through conversation with AI.

Getting Started

1. Sign up and generate an API Key from Dashboard > Settings > API Keys.
2. Copy the configuration below for your AI client and paste it into the settings file.
3. Tell your AI “Create a video for me” and you're done!

Connection Info

Item	Value
MCP Endpoint	https://autoagens.com/api/mcp
Transport	Streamable HTTP
Auth	Authorization: Bearer ak_YOUR_API_KEY

---

Connect with Claude Desktop

Open the Claude Desktop config file and add the following:

• Mac: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "autoagens-video": {
      "type": "url",
      "url": "https://autoagens.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ak_YOUR_API_KEY"
      }
    }
  }
}

Replace ak_YOUR_API_KEY with your actual API key. Save and restart Claude Desktop to connect.

Connect with Claude Code (CLI)

Run this command in your terminal:

claude mcp add autoagens-video \
  --transport http \
  https://autoagens.com/api/mcp \
  --header "Authorization: Bearer ak_YOUR_API_KEY"

Connect with Cursor

Create a .cursor/mcp.json file in your project root:

{
  "mcpServers": {
    "autoagens-video": {
      "type": "url",
      "url": "https://autoagens.com/api/mcp",
      "headers": {
        "Authorization": "Bearer ak_YOUR_API_KEY"
      }
    }
  }
}

Windsurf / Other MCP Clients

Any client supporting MCP Streamable HTTP can connect:

URL:    https://autoagens.com/api/mcp
Method: POST
Header: Authorization: Bearer ak_YOUR_API_KEY
Header: Content-Type: application/json
Header: Accept: application/json, text/event-stream

---

Verify Connection

After connecting, try telling your AI:

• “Read the video-guide prompt” — to see the full usage guide
• “List available tools” — to see all 11 tools

---

Available Tools

Workflow Management

Tool	Description
create_workflow	Create a new video workflow
save_workflow	Save nodes and edges to a workflow
list_workflows	List your workflows

Assets & Images

Tool	Description	Cost
upload_asset	Upload an external image URL or base64 as a stable asset. Reuse the returned assetId across tools	Free
generate_image	AI image generation (text/image based). Automatically saved as asset with assetId	100 coins

Video Generation

Tool	Description	Cost
create_clip	Create video clip (async). Prompt required for ALL modes. Supports video continuation for scene continuity	2,500~7,500 coins
compose_video	Compose clips into final video (async)	500 coins

Status & Frame Extraction

Tool	Description
get_job_status	Check clip/compose job status
extract_frame	Extract a frame from a completed clip. For multi-scene chaining (sync, 5-15s)

Templates

Tool	Description
list_templates	List public templates
use_template	Copy template to your workflows

Supported AI Video Models

Model	Modes	Duration	Features
veo3-fast	T2V	5-8s	Google Veo 3 fast, seed/negative_prompt
veo3	T2V	5-8s	Google Veo 3, highest quality
kling-v3-pro	I2V, IT2V	3-15s	Character refs/endImage/negative_prompt/video continuity
kling-o3	I2V, IT2V	3-15s	High quality, character refs/endImage/negative_prompt/video continuity
kling-v2.6-pro	I2V, IT2V	5 or 10s	endImage/negative_prompt
ltx-2.3	I2V, IT2V	6-20s (even only)	Fast generation, endImage

Advanced Options (model-specific)

Feature	Description	Supported Models
negativePrompt	Specify unwanted elements	veo3, kling models
seed	Reproduce identical results	veo3-fast, veo3
characterReferences	Character consistency (frontal + reference images)	kling-v3-pro, kling-o3
endImageUrl	Specify end frame image	kling, ltx models

Video Continuity Options

For multi-scene production, reference the previous scene's video to enhance motion, camera, and scene layout continuity.

Parameter	Description
videoUrl / referenceVideoUrl	Previous clip video URL for motion/camera continuity reference
previousClipTaskId	Completed previous clip taskId. Server auto-resolves its videoUrl
continueMotion	Continue motion state from previous clip (default true with reference video)
preserveCamera	Preserve camera direction/movement (default true with reference video)
preserveSceneLayout	Preserve scene layout/blocking (default true with reference video)

The backend currently processes continuity via frame extraction + prompt augmentation. Korean prompts are automatically translated to English before being sent to the model.

---

Video Creation Flow

Just tell your AI “Create a 30-second ad video” and it handles everything automatically. Under the hood:

1. Create workflow — Start with create_workflow
2. Prepare reference images — Use upload_asset to import external images as stable assets (free)
3. Generate images (optional) — Create source images with generate_image (100 coins)
4. Create clips — create_clip for each scene. Prompt required for all modes (async, 1-3 min)
5. Check status — Poll get_job_status every 3-5 seconds
6. Compose video — Combine all clips with compose_video (2-5 min)
7. Download result — Get the final video URL

Multi-Scene Chaining

Passing just a single frame creates visual continuity but may lack motion/camera/story continuity. For best results, pass both the end frame and the previous video reference.

1. Scene 1 create_clip → wait for completion
2. extract_frame(taskId: "scene1_taskId", position: "end") → get frameUrl
3. get_job_status(jobId: "scene1_taskId") → get previous scene videoUrl
4. Scene 2 create_clip with:
   • imageUrl: frameUrl — visual continuity
   • referenceVideoUrl: "scene1_videoUrl" — motion/camera continuity
   • prompt: "next scene description" — required
5. Repeat for all scenes, then compose_video

Or use previousClipTaskId for automatic videoUrl resolution:

create_clip({
  imageUrl: "frameUrl",
  previousClipTaskId: "scene1_taskId",
  prompt: "next scene description",
  model: "kling-v3-pro",
  duration: 5,
  clipId: "scene-02"
})

Pricing

Action	Cost
Upload asset	Free
Generate image	100 coins
Extract frame	Free
Create clip (5s)	2,500 coins
Create clip (10s)	5,000 coins
Create clip (15s)	7,500 coins
Compose video	500 coins

Failed jobs are automatically refunded. Admin users can use all tools for free.

FAQ

Where do I get an API Key?

Generate one from Dashboard > Settings > API Keys. Keys start with ak_.

Which AI clients are supported?

Any client that supports MCP Streamable HTTP. Tested with Claude Desktop, Claude Code, Cursor, and Windsurf.

What is the difference between upload_asset and generate_image?

upload_asset stores an existing image on Autoagens (free). generate_image creates a new AI image and auto-saves it (100 coins). Both return an assetId.

What is extract_frame for?

It extracts a frame from a completed video clip. Used in multi-scene chaining to pass the last frame of the previous scene as the start image for the next scene.

How do I create smooth scene transitions?

Use extract_frame to get the end frame, then pass both imageUrl (the frame) and referenceVideoUrl or previousClipTaskId (the video) to the next create_clip. This provides both visual and motion/camera continuity.

Can I use Korean prompts?

Yes, both Korean and English are supported. Korean prompts are automatically translated to English before being sent to the model.

Where can I buy coins?

Visit Dashboard > Coins to purchase coins.