The open-source MCP video server for Claude Code

What is Model Context Protocol?

MCP is an open standard from Anthropic for connecting AI assistants to external tools, data sources and prompts. An MCP server exposes capabilities; an MCP client (Claude Code, Claude Desktop, Cursor) discovers and invokes them. The protocol is transport-agnostic — ViralMint mounts the server on HTTP Streamable at the loopback address the desktop binds.

The practical effect: instead of clicking through ViralMint's UI for every step, you tell Claude what you want and it calls the right tools in sequence. Multi-step workflows that used to need 6–9 manual clicks collapse into one prompt.

Setup in 60 seconds (Claude Code)

1. Download ViralMint from viralmint.net/#download (macOS / Windows / Linux).
2. Open the /mcp page in the desktop app sidebar. The Setup tab shows your auto-generated bearer token.
3. Run the one-liner: claude mcp add --transport http viralmint http://127.0.0.1:16888/mcp --header "Authorization: Bearer <your-token>"
4. Restart Claude Code and try the example prompt below.

Config snippet (Claude Desktop / Cursor)

# .mcp.json (Claude Code / Cursor)
{
  "mcpServers": {
    "viralmint": {
      "transport": {
        "type": "http",
        "url": "http://127.0.0.1:16888/mcp",
        "headers": {
          "Authorization": "Bearer <token-from-desktop-app>"
        }
      }
    }
  }
}

The /mcp page → Setup tab generates this snippet pre-filled with your bearer token. Click the copy button and paste into .mcp.json or the equivalent client config.

What's in the 136 tools, by workflow stage

Example: multi-step creator pipeline in one prompt

# Example prompt to Claude Code with ViralMint MCP installed

> Scout trending TikTok videos in the "productivity hacks" niche.
> Download the top 3 outliers. Transcribe them with Whisper.
> Then generate a fresh Smart Video that uses a similar hook
> pattern but covers my topic: "morning routines for night owls".

# Under the hood, Claude calls:
#   scout_trending(niche="productivity hacks", platforms=["tiktok"])
#   wait_for_job(job_id)
#   download_video(ids=[...top 3 outlier ids])
#   wait_for_job for each download
#   analyze_downloaded(ids=[...])
#   wait_for_job for each analysis
#   generate_from_downloaded(downloaded_video_id=..., topic="morning routines for night owls")
#   wait_for_job, then output_path is your finished mp4.

The full chain takes ~3–8 minutes wall-time (depending on Whisper model + AI video tier) and costs $0.05–$0.15 in cloud credits. Claude waits on each wait_for_job before continuing, so the prompt flow is fully autonomous.

Architecture notes

• Loopback-only by default. The MCP endpoint binds to 127.0.0.1:16888. Not exposed to the network unless you explicitly front it with a reverse proxy. Bearer token is the install-binding signal so one Claude Code config maps to one ViralMint install.
• HTTP-to-self. MCP tools call the same /api/* routes the web UI uses, via httpx loopback. ~0.5–1 ms per tool call. Trade-off: middleware (CSRF, auth, rate-limits) runs once and is shared between UI and MCP, so no surface can bypass a check added for the browser.
• Real-time activity streaming. Every MCP call broadcasts a mcp_activity WebSocket event so the /mcp Activity tab updates within milliseconds. Useful for debugging chains and for billing transparency.
• Cost preview built-in. tool_cost_estimate(tool, args) returns {cost_usd, model, notes} for any tool without invoking it, so Claude can warn before spending. Covers the 5 expensive primitives plus the create_clip_sequence macro.

How does this compare to other MCP servers?

Most MCP servers expose read-only data sources (filesystem, GitHub, Slack) or single-shot actions. ViralMint is the only end-to-end creator pipeline MCP server we know of — it's the difference between "Claude can read your Slack messages" and "Claude can run a 6-stage video production pipeline and hand you the finished mp4". The differentiation is the wait_for_job orchestration helper, the macros that collapse multi-call workflows, the cost estimation tool, and the bundled Chromium for headless browser automation.