Connect ngram over MCP

Overview

The Model Context Protocol is an open standard for connecting AI assistants to external tools. ngram exposes every public endpoint as an MCP tool with a typed schema the agent reads at connect time, so there is no glue code to write. There are two ways to connect: the hosted HTTP endpoint (recommended) or a local stdio process.

Connect to the hosted server

Point your MCP client at https://mcp.ngram.com and authenticate with Authorization: Bearer ngs_…. That is the entire setup — nothing to install. Create a key first in Settings → API keys.

Run it locally over stdio

Desktop clients that launch a local process can run the official package with npx and read the key from the environment. The package @ngramdata/mcp-server ships a single executable named ngram-mcp-server. It is distributed through GitHub Packages, so most teams prefer the hosted endpoint above unless they specifically need a local stdio process.

{
  "mcpServers": {
    "ngram": {
      "command": "npx",
      "args": ["-y", "--package=@ngramdata/mcp-server", "ngram-mcp-server"],
      "env": {
        "NGRAM_API_KEY": "ngs_your_key"
      }
    }
  }
}

Set NGRAM_API_KEY before launching. To target a staging or self-hosted ngram, override NGRAM_PUBLIC_API_BASE_URL (it defaults to https://www.ngram.com).

Tool catalog

Every ngram endpoint maps to a discrete tool. The agent reads the typed schema for each at connect time, so it can pick the right one without a docs lookup mid-conversation.

Account

• get_account — return the authenticated ngram account.
• get_credits_remaining — return the credit balance and plan.

Catalog

• list_voices, list_styles, list_aspect_ratios, list_durations — the options a render can use.
• list_animation_modes, list_scenarios, list_brand_kits — resolve modes, presets and your brand kits.

Videos

• create_video_from_text and create_video_from_url — start a render from a prompt or a source URL.
• prepare_video then create_video — preview the resolved settings, then confirm and generate.
• get_video_status, list_videos, cancel_video — track, list and stop renders.

Images

• create_image, edit_image, get_image_status, list_images — generate and edit images the same way.

Example prompts

Once connected, just ask in plain language — the agent picks the right tool:

• "Make a 30-second product teaser for our pricing page and tell me when it's done."
• "How many credits do I have left?"
• "List my last five videos and their status."