Use it in your AI assistant (MCP)

One MCP server covers every dataset. It uses the same key as the API, and its calls count toward the same limits.

Server address: https://api.heimdallresearch.com/mcp (streamable HTTP)

GET /v1/mcp/setup returns this configuration for your account, ready to copy.

Claude Code

claude mcp add --transport http heimdall https://api.heimdallresearch.com/mcp \
  --header "Authorization: Bearer $HEIMDALL_API_KEY"

Cursor, and other clients that use a JSON config

{
  "mcpServers": {
    "heimdall": {
      "url": "https://api.heimdallresearch.com/mcp",
      "headers": { "Authorization": "Bearer ${HEIMDALL_API_KEY}" }
    }
  }
}

ChatGPT

Settings, Connectors, Add custom connector. Paste the server address and set the Authorization header to Bearer followed by your key.

Tools

Tool	Parameters	Ask it something like
list_datasets	none	“What data does Heimdall Research publish?”
describe_dataset	slug	“How is lake area measured, and what are the limits?”
get_storm	event_id, limit (1–100 counties, default 10)	“How long did counties in Puerto Rico stay dark after Maria (AL152017)?”
get_lake_series	lake_id, cadence (monthly, or weekly on a paid plan)	“Walker Lake area each month since 2019”
get_arctic_gate	gate_id	“When did the Bering Strait gate open last season?”
get_download_link	dataset, optional file name	“Give me the full storm file”
check_usage	none	“How many calls do I have left today?”

Each answer carries the dataset version and citation. Data that is not released yet returns a short note saying when it is planned.