# Connect an AI agent with MCP

> Drive Katechon Arc from any MCP client. The remote server exposes the same production workflow as the studio — projects, scripts, references, storyboards, video and voice — to your own agent.

URL: https://arc.katechon.technology/developers/mcp
Updated: 2026-10-17

## Connection details


| Setting | Value |
| --- | --- |
| Endpoint | https://arc.katechon.technology/mcp/ |
| Transport | Streamable HTTP |
| Authentication | Authorization: Bearer YOUR_API_KEY (a personal key with the arc- prefix) |
| Tools | 38 |

## 1. Create an API key

Sign in and open [Settings → API keys](https://arc.katechon.technology/app/settings?section=api-keys&create=true). Every signed-in user can create a personal key. It acts with that account's permissions and project access, and projects created over MCP belong to that account. The full key is shown once; you can list and revoke your keys at any time. Generations started over MCP use that account's credits, as they do in the studio.

## 2. Add the server to your client

Claude Code:

```
claude mcp add --transport http katechon-arc https://arc.katechon.technology/mcp/ --header "Authorization: Bearer YOUR_API_KEY"
```

Cursor and other clients that read an mcpServers JSON file:

```
{
  "mcpServers": {
    "katechon-arc": {
      "url": "https://arc.katechon.technology/mcp/",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}
```

Setup is working when the list_projects tool returns a projects list (an empty list is fine). Agents can also follow the step-by-step [agent installation guide](https://arc.katechon.technology/agent-installation-guide.md).

## How generation works over MCP

Media tools submit durable generation batches and return immediately. Poll get_generation_batch as advised by poll_after_seconds until the batch is done, then read which items succeeded, failed or were blocked. Text tools that support a dry run return the prompt without starting a batch.

## Tools

### Projects and source material

- list_projects — List projects
- create_project — Create project
- upload_source — Upload source
- list_source_files — List source files
- get_source_text — Read source text
- get_project_content — Read project content
- patch_project — Edit project assets
- rename_asset — Rename asset
- list_project_files — List project files
- read_project_file — Read project file
- retry_project_migration — Retry data upgrade

### Planning and scripts

- get_workflow_plan — Get complete workflow plan
- plan_episodes — Plan episodes
- reset_episode_planning — Reset episode planning
- generate_script_plan — Generate script plan
- get_script_plan_content — Read script plan content
- complete_script_plan_rebuild — Complete script plan rebuild
- confirm_script_review — Complete content confirmation
- generate_episode_script — Generate script
- get_episode_script — Read episode script
- patch_episode_script — Edit script
- patch_episode_meta — Edit episode title
- open_draft — Open draft
- patch_draft — Edit draft
- promote_draft — Promote draft
- discard_draft — Discard draft

### References, storyboards, video and voice

- list_pending_assets — List pending assets
- complete_asset_inventory — Complete asset inventory
- generate_assets — Generate assets
- edit_images — Edit images
- generate_storyboards — Generate storyboards
- generate_grid — Generate multi-grid storyboard
- get_video_capabilities — Query video model capabilities
- generate_videos — Generate videos
- generate_narration_audio — Generate narration audio
- get_prompt_preview — Preview final prompts

### Generation batches

- get_generation_batch — Get generation batch
- cancel_generation_batch — Cancel generation batch

## Limits and expectations

- API keys are bearer credentials. Keep them out of shared repositories and prompts.
- Requests are refused when the account cannot cover their estimated credits.
- There is no OAuth sign-in flow; clients must send the API key header.