Mirvio Cloud Video Tools

npx add-mcp 'https://mirvio.ai/api/mcp'

1. Create a token

1. Sign in and open Account.
2. Under “MCP access for AI clients”, enter a name and select Create token.
3. Copy the token. Mirvio shows it only once and stores only a hash.
4. To stop access, select Revoke. The next request with that token fails.

The token goes in your client configuration, not in a chat message. Do not share it. Each token can only use your own uploads, jobs, and allowance.

2. Add the server to your client

Endpoint: https://mirvio.ai/api/mcp (Streamable HTTP, stateless). Put the token in an environment variable named MIRVIO_MCP_TOKEN.

Claude Code

claude mcp add --transport http mirvio https://mirvio.ai/api/mcp \
  --header "Authorization: Bearer $MIRVIO_MCP_TOKEN"

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "mirvio": {
      "url": "https://mirvio.ai/api/mcp",
      "headers": { "Authorization": "Bearer ${env:MIRVIO_MCP_TOKEN}" }
    }
  }
}

VS Code (.vscode/mcp.json)

{
  "inputs": [
    { "type": "promptString", "id": "mirvio-token", "description": "Mirvio MCP token", "password": true }
  ],
  "servers": {
    "mirvio": {
      "type": "http",
      "url": "https://mirvio.ai/api/mcp",
      "headers": { "Authorization": "Bearer ${input:mirvio-token}" }
    }
  }
}

OpenAI Codex CLI (~/.codex/config.toml)

[mcp_servers.mirvio]
url = "https://mirvio.ai/api/mcp"
bearer_token_env_var = "MIRVIO_MCP_TOKEN"

Claude Desktop and other stdio-only clients

Use the mcp-remote bridge (requires Node.js):

{
  "mcpServers": {
    "mirvio": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mirvio.ai/api/mcp",
               "--header", "Authorization:${AUTH_HEADER}"],
      "env": { "AUTH_HEADER": "Bearer paste-your-token-here" }
    }
  }
}

3. Run a job

1. Call get_capabilities for supported operations and recipes. Then call get_account to see your remaining cloud outputs.
2. create_upload returns a signed address and a curl command. The client runs the command on your computer. The upload returns a receipt.
3. confirm_upload takes that receipt.
4. submit_jobs queues the jobs. Each Creative Mesh variant uses one cloud output. Other operations use one each. The client must set acknowledgeVideoJobs to the exact output count (up to 40), or the request fails without a charge. A repeated requestKey returns the first batch without a second charge.
5. get_job_status shows progress. Jobs run asynchronously, usually within a few minutes. A job that fails, or a queued job that does not finish within 6 hours, returns its cloud output. A finished job keeps its charge, even if its result expires before you download it.
6. get_download_url returns a link that works for 5 minutes. Results expire after 24 hours.

Tools and costs

Tool	What it does	Cost
get_capabilities	Supported operations, limits, and effect recipes	Free
get_account	Plan, remaining cloud outputs, and costs	Free
create_upload	Signed upload address and a curl command for one file	Free
confirm_upload	Confirm the upload with its receipt	Free
submit_jobs	Queue 1–10 video or audio operations	1 per variant or operation
get_job_status	Status of each job in a batch	Free
get_download_url	Signed 5-minute download link for a finished job	Free

Reframe targets 9:16, 16:9, 1:1, or 4:5 with a crop or a blurred background. Compress targets a file size from 100 KB to 1 GB. Metadata replaces device and capture tags.

Try these workflows

• “Add an upward RGB pixel scan to the first seven seconds of this video. Make one Creative Mesh variant.”
• “Use Side Voice + Center Layer with this video and my separate center audio track.”
• “Join these three clips, extract twelve thumbnails, and make a 9:16 version with a blurred background.”

Upload and confirm every source file first. The audio effect requires a separate center track and explicit agreement: the base voice cancels when played in mono. The client can read starting settings from get_capabilities. These reproduce effect types; they do not recover the exact settings of another video.

Supported operation names: reframe, compress, metadata,creative_mesh, audio_mesh, thumbnails,audio_split, video_join, and audio_join. A join uses confirmed upload IDs in the desired order. Download a completed result and upload it again when you want to use it as the source for a later step.

The same jobs in the web tools

Each MCP job runs the same cloud tool as the web form. Read what each tool changes before you connect a client.

Limits

• Authentication uses a personal bearer token. Clients that connect only through OAuth, such as claude.ai web connectors and ChatGPT connectors, are not supported yet.
• The client must be able to run curl (or another HTTP request) to upload the file.
• Upload limits are the same as the web tools: 2 GB per file and 100 files or 10 GB a day; 250 MB per file and 20 files on the free trial.
• Up to 10 active tokens per account.