Agent video downloader

Turn public video links into an agent capability.

Fastdown lets users ask an AI agent to parse, inspect or download a public short-video link. The agent gets a clear MCP tool contract, and the user gets local files or structured media data.

Who should use this

The best early users are already using agents, terminals, content automation, clipping workflows or internal research tooling.

Creators

Material collection

Ask an agent to save public links into a predictable folder before editing or review.

Developers

Automation scripts

Use JSON output and quota commands in shell scripts, workers or internal dashboards.

Teams

Repeatable workflow

Free, Pro and Max quotas make daily parsing volume explicit and easy to manage.

Agent decision rules

These rules are mirrored in /llms.txt and the agent manifest so crawlers and assistants can route requests correctly.

  1. 1If the user asks to save or download media, call download_video.
  2. 2If the user asks for media URLs or metadata, call parse_video.
  3. 3If the user asks about limits, call quota.

What a parse returns

A parse is not a download button. It is structured output: every rendition, whether it carries a watermark, and how much quota is left.

$ fast-download parse "https://v.douyin.com/xxxx/" --json | jq '.media[0]'
{
  "quality": "1080p",
  "ext": "mp4",
  "watermark": false,
  "url": "https://..."
}

Coverage and limits

Stated plainly, because a tool that hides its edges is a tool you cannot build on.

CapabilityStatusNotes
Platforms detected automatically7 sourcesTikTok, Douyin, Xiaohongshu, YouTube Shorts, Reels, Facebook, X
Output formatsJSON or file--json for pipelines, plain download for humans
Agent clients supported6Codex, Claude Code, Cursor, VS Code, OpenCode, Windsurf
ConcurrencySequential per accountKeeps quota accounting exact
Windows supportYesSingle binary plus a PowerShell installer
Self-hosting the APINoThe parsing upstream is licensed, the client is not

Wire it into a client (optional)

Skip this section if you only ever want the terminal.

// Claude Code
$ claude mcp add fast-download -- fast-download mcp --stdio

// Cursor — .cursor/mcp.json
{
  "mcpServers": {
    "fast-download": { "command": "fast-download", "args": ["mcp", "--stdio"] }
  }
}

Boundaries and compliance

Four constraints that make the tool safe to put inside an automated workflow.

Questions people actually ask

What counts as an agent video downloader?

A tool an autonomous agent can call without a human in the loop: a stable interface, machine-readable output and predictable quota. Fastdown ships that as a CLI plus an MCP server.

Can an agent decide by itself which link to download?

It can, but you should scope it. The MCP tools accept a URL and nothing else — no wildcard crawling, no profile scraping, no search. The agent acts on links a human or a pipeline already chose.

Does it work in CI or cron?

Yes. fast-download download URL --json is a single process with a non-zero exit code on failure, which is all a cron job needs.

Is there a rate limit?

The free tier allows 50 successful parses per day; Pro is 1,000 and Max is 3,000. Limits reset daily and failed parses are never counted.

What stays predictable

Agent workflows break on surprises, so these three properties are the product.

  1. 1One schema, seven platforms — The JSON shape does not change between TikTok, Douyin, Xiaohongshu and the rest, so the agent's parsing logic is written once.
  2. 2Quota is readable — The quota tool is callable at any point, which lets an agent plan a batch instead of discovering the limit at parse 51.
  3. 3Exit codes are meaningful — Zero means a file exists at the returned path. Anything else means no file was written — never a half-downloaded artifact.