How it works
- A server publishes tools. Each tool has a name, a description and typed inputs, such as create_project or generate_videos.
- A client connects. An AI assistant or coding agent is configured with the server's URL and a credential.
- The model chooses tools. When your request needs one, the assistant calls it with arguments and reads the result.
- Long jobs run asynchronously. Video takes minutes, so good servers return a job or batch id that the agent checks later.
An example: what Arc's server exposes
- Projects and source material (11 tools): list_projects, create_project, upload_source, list_source_files, …
- Planning and scripts (15 tools): get_workflow_plan, plan_episodes, reset_episode_planning, generate_script_plan, …
- References, storyboards, video and voice (10 tools): list_pending_assets, complete_asset_inventory, generate_assets, edit_images, …
- Generation batches (2 tools): get_generation_batch, cancel_generation_batch
The endpoint is https://arc.katechon.technology/mcp/, authenticated with a personal API key. Details on the developers page.
MCP versus an API or an agent
| Term | What it is |
|---|---|
| API | Endpoints a programmer calls from code. |
| MCP server | Tools described so an AI model can discover and call them without custom code. |
| AI agent | A model that plans steps and calls tools, often through MCP, to finish a task. |
Other video tools with MCP servers
Higgsfield: Official MCP server at mcp.higgsfield.ai for Claude, ChatGPT and other clients, with prebuilt ad and production workflows. Source: Higgsfield (higgsfield.ai/mcp), checked 2026-10-17
Runway: Runway MCP for ChatGPT, Claude, Cursor and others, using account credits; a Dev MCP for the API. Source: Runway (runway.com/mcp), checked 2026-10-17
What to watch for
- Credentials: an MCP key acts as your account. Keep it private and revoke it when unused.
- Spend: an agent can start many paid generations in one turn. Ask it to estimate first.
- Review: keep a human checkpoint before the expensive step, the way a script plan is confirmed before episode scripts.
Where Katechon Arc fits
Arc runs a remote MCP server with 38 tools, so an agent can drive a whole production with the same estimates and review gates as the studio.
Arc's server requires a paid account and an API key; it is not a free generation endpoint and does not offer models beyond Seedance 2.0, Seedance 2.0 Fast, Seedance 2.0 Mini.
People also ask
What does MCP stand for in AI?
Model Context Protocol, an open standard for connecting AI assistants to external tools and data.
Is MCP the same as an AI agent?
No. MCP is the connection standard; an agent is the model that decides which tools to call, often through MCP servers.
Related answers
Sources
- Higgsfield: higgsfield.ai/mcp, checked 2026-10-17
- Runway: runway.com/mcp, checked 2026-10-17