MCP

MCP is the protocol an AI assistant uses to reach external tools. @airi.live/mcp is ours: it exposes your account as 32 tools that Claude, Cursor or any compatible client can call — upload a file, quote an encode, start it, wait for it, create a player and hand you the embed.

It runs on your machine, not ours. Your credential never leaves it, and an upload's bytes go from disk straight to the bucket — through neither our server nor the assistant.

First mint a credential in Developers, ticking the permissions you want the assistant to have. Then one command:

claude mcp add airi --env AIRI_KMS_TOKEN=kms_live_... -- npx -y @airi.live/mcp

For Claude Desktop, Cursor, Windsurf and anything else configured with JSON:

{
  "mcpServers": {
    "airi": {
      "command": "npx",
      "args": ["-y", "@airi.live/mcp"],
      "env": { "AIRI_KMS_TOKEN": "kms_live_..." }
    }
  }
}

Node 20 or newer. Nothing to install — npx fetches it on first run and keeps it current.

Both forms work and the flag wins over the variable, but the token belongs in the env block: an argument shows up in your machine's process list and an environment variable does not. npx @airi.live/mcp --help lists everything.

From there you just ask. For example: *"upload ~/Videos/keynote.mp4, encode it at 1080p and 720p with DRM, and give me an embed to paste into our site"*.

airi_storage_upload   path=~/Videos/keynote.mp4        → file id
airi_encoding_quote   heights=[1080,720] duration=600 → $0.16
airi_encoding_start   package=hls+dash drm=true       → job id
airi_encoding_job     wait=true                       → completed
airi_player_create    name="Site"                     → player id
airi_player_embed     file=<id>                       → signed URL + <iframe>

Three of them make more than one call, because the API is built for browsers and an assistant is not one. airi_storage_upload takes a path and returns a sealed file — declare, multipart upload, ETag collection and complete, all inside the tool. airi_encoding_job with wait blocks until the job finishes, so an encode is one step rather than a poll loop. And airi_player_embed returns the ready <iframe> alongside the resolved player.

The last two are an escape hatch: they read the API reference live and run whatever they find. A new endpoint is available the day it ships, without waiting for us to update the package.

You are giving a language model access to your account. Two things the server refuses to do even if asked, and they are worth knowing before you install it.

Everything else is reachable, because it is the job. Deletes and encodes carry the protocol's destructiveHint, so your client can ask before running them, and the descriptions say plainly what cannot be undone: a contentId is permanent at the DRM provider, a deleted file has no undo, and cancelling an encode stops the billing rather than the encoder.

The credential is the real boundary: the server cannot grant what your token does not have. Mint one for the assistant rather than reusing your own, ticking only the permissions the work needs.

A read-only assistant is a credential minted without write access: it can look at everything and change nothing. Revoking it from Developers cuts it off on the next call, not a minute later.

Errors reach the assistant with their code and a hint about what to do, so it usually corrects itself and tells you. If not, your MCP client's logs carry whatever the server wrote.