KPainter

Use KPainter's remote MCP server to create source-grounded videos, slides, and interactive knowledge content from your own materials.

View on kpainter.ai

Documentation

Quickstart

This page shows the minimum MCP setup that gets KPainter running inside an MCP-compatible client.

Step 1: Get an API Key

KPainter MCP uses the user's own API key.

Before you configure any client:

  1. sign up or log in to KPainter
  2. open the API Key page
  3. activate and copy your API key
  4. keep it ready for MCP client config

Step 2: Add a Local stdio MCP Server

For local MCP clients, stdio is the easiest path.

If kp-mcp is already available in your environment, a standard config looks like this:

Copy

{  "mcpServers": {    "kpainter": {      "command": "kp-mcp",      "env": {        "KGP_API_KEY": "<your_api_key>"      }    }  }}

If you are running from the kgp-service source tree instead, use:

Copy

{  "mcpServers": {    "kpainter": {      "command": "python",      "args": ["-m", "src.mcp_server"],      "cwd": "/path/to/kgp-service",      "env": {        "KGP_API_KEY": "<your_api_key>"      }    }  }}

Step 3: Validate the Key First

The CLI binary is kp-mcp.

The MCP tool names now use the kp_* prefix, so start with tools such as kp_me and kp_get_catalog.

Once the MCP server is connected, start with:

  • kp_me

Use it to confirm:

  • the key is valid
  • it belongs to the intended user
  • it can manage creations

Do not skip this and jump straight into creation.

Step 4: Read the Catalog

Call:

  • kp_get_catalog

Use it before choosing:

  • type
  • aspect_ratio
  • output_quality
  • voice_id
  • style_id
  • duration_seconds
  • scene_count
  • image_provider / image_model / image_quality
  • video_model / video_task / video_resolution / video_duration_seconds

Do not guess these values.

Step 5: Create the First Result

A first happy-path workflow usually looks like this:

  1. kp_me
  2. kp_get_catalog
  3. kp_create_creation
  4. kp_get_job_status
  5. kp_get_creation

For example, you can ask your MCP client to create something like:

Create a Read Aloud Video in English that explains how MCP works in 6 pages.

The client should then map that request to the official type and parameters returned by kp_get_catalog.

The MCP product pages show the exact flat arguments accepted by kp_create_creation. Do not copy the nested image_generation or video_generation shapes from the REST OpenAPI into an MCP tool call.

Step 6: Read the Final Detail

When the job succeeds, read:

  • main_url
  • artifacts
  • scenes
  • editable_actions

from kp_get_creation.

This is the main place to read the finished result.

If You Need Scene-Level Editing

For supported types, inspect editable_actions and scenes first, then use:

  • kp_edit_creation

Current scene-edit actions are:

  • update_scene_narration
  • regenerate_scene

Treat edits as async too:

  1. kp_edit_creation
  2. kp_get_job_status
  3. kp_get_creation

Read Next

  • Authentication
    understand how api_key and default-key fallback work
  • Transport Modes
    see when to use stdio or Streamable HTTP
  • Available Tools
    browse the current tools, resources, and prompts