Claude Code generates videos: Jantaris MCP step by step

Claude Code can use Jantaris MCP to create media and save it beside your code. The model running your conversation and the model generating a file have separate roles. Adding MCP does not automatically replace your editor's chat model.

1. Create a dedicated key

• Open API keys, create a key for Claude Code and set a monthly limit.
• Copy the key when it is shown. Keep it out of your repository and task prompt.
• Generation requires an active service and sufficient balance. Connecting MCP does not start a paid task.

2. Connect Claude Code

claude mcp add --transport http jantaris https://jantaris.com/mcp --header "Authorization: Bearer sk-jt-..."
claude mcp list

Replace sk-jt-... with your key. For Codex and Cursor, set JANTARIS_API_KEY in the environment that launches the app. PowerShell uses $env:JANTARIS_API_KEY="sk-jt-..."; macOS and Linux use export JANTARIS_API_KEY="sk-jt-...". Restart your editor after changing its environment. Cursor project configuration belongs in .cursor/mcp.json. See the MCP setup guide for all installation options.

3. Ask for a quote before generation

Prepare three vertical coffee shop videos. First check Jantaris MCP models and required inputs. Quote every variant with estimate_cost, show the total and wait for my approval. Only then generate with confirm_token. Download results to public/generated/ and verify their type and dimensions. Do not overwrite existing files or publish automatically.

• Use search_models and get_model to inspect available inputs.
• Call estimate_cost for every planned file and show the total.
• After approval, pass each matching confirm_token to generate. Read progress with wait_for_task or get_task.
• A different prompt, model or resolution requires a fresh quote and approval. Confirmation tokens expire.

Start with Seedance 2.5. Current unit rate: from $0.261 per second. This is not a full project quote. Check settings and quantity in the calculator and pricing.

4. Save and review the files

• Wait for succeeded. A queued task has no completed output yet.
• Download URLs from output to your project. Links expire; do not hard-code them into your website.
• Verify each local file, its dimensions and the complete visual result before use.
• Keep the model, prompt, settings and cost. A successful output you choose not to use is still billable. Failed generation is free.

Troubleshooting

• Connection failure: check the /mcp URL, key status and environment visible to the editor process.
• Authorization or balance error: check the key limit and account balance.
• Invalid input: read get_model again instead of copying another model's settings.
• Interrupted response: inspect the saved task ID before submitting another generation.

Back to AI agents. Add AGENTS.md to a project for API instructions.

FAQ

Does the agent need a separate balance?

No. MCP and Studio use the same Jantaris account and balance. Your editor or conversation model service may have separate charges.

Can the agent save a file itself?

Yes, if it has network and project file access. Specify the output folder and ask it to verify the saved file. A URL alone is not a local file.

Does getting a quote start generation?

No. estimate_cost returns a quote and confirmation token. generate starts the task. Tell the agent to wait for approval and set a key spending limit.