@pondlog/mcp-npn

A place-aware USA National Phenology Network MCP server. Gives any MCP-aware AI client (Claude Desktop, Cursor, Continue, custom agents) eight tools for plant and animal phenology data: species catalog, stations, observations, site-level phenometrics, and a place-aware "what's active near me" shortcut.

No API key required. NPN is a public scientific dataset.

Part of pondlog, a toolkit for stitching together public nature APIs (iNaturalist, eBird, USGS, NPN, NOAA, SunCalc).

What is phenology?

Phenology is the seasonal timing of biological events: first leaf, first bloom, first call, first migration. NPN volunteers and scientists record these events at thousands of stations across the US, building a multi-decade dataset used for tracking climate change, planning restoration, and understanding regional ecology.

Install / run

The server is published to npm and runs via npx (no install step):

npx -y @pondlog/mcp-npn

It speaks MCP over stdio. The server is internally rate-limited (1 req/sec sustained, bursts to 5). Be polite; NPN doesn't publish a formal limit.

Configure

Claude Desktop

Edit ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "pondlog-npn": {
      "command": "npx",
      "args": ["-y", "@pondlog/mcp-npn"]
    }
  }
}

Restart Claude Desktop. The eight tools will appear in the slash menu.

Cursor

Edit ~/.cursor/mcp.json:

{
  "mcpServers": {
    "pondlog-npn": {
      "command": "npx",
      "args": ["-y", "@pondlog/mcp-npn"]
    }
  }
}

MCP Inspector (debug)

npx @modelcontextprotocol/inspector npx -y @pondlog/mcp-npn

Tools

All tools are read-only. Coordinates use WGS84 decimal degrees. State codes are 2-letter US postal codes (WA, CA, NY).

Tool	What it does
search_species	Filter the ~1,900-species NPN catalog by name, genus, or kingdom. Returns NPN species_ids you can pass to other tools.
get_stations_in_state	List NPN stations in a US state. Returns station_ids, names, coordinates.
get_station_count_by_state	Quick map of NPN coverage: how many stations per state.
get_stations_with_species	Find stations that have observed specific species.
get_stations_by_location	Stations inside a WKT polygon. Use get_active_phenology_nearby for the lat/lng/radius shortcut.
get_observations	Raw status/intensity records (was phenophase 'yes'/'no' on date). Requires year + ≥1 narrowing filter. Bandwidth-heavy.
get_site_level_data	Per-site phenometric aggregates: mean first/last 'yes' date for each (site, species, phenophase). Best granularity for "when does X bloom here?"
get_active_phenology_nearby	Place-aware shortcut: WKT bbox, stations, haversine, site-level data, sorted by most-recent activity. The headline tool.

Key gotchas

• Coverage is patchy. Eastern US and Arizona are dense; the Pacific Northwest is sparser. Use get_station_count_by_state first if you're not sure.
• NPN stations are species-specific. A station near you might track only invasive plants or only bird arrivals. Cross-reference with get_stations_with_species.
• Data lags by a year. Phenometric records for the most recent year are still being ingested. Default years_back: 2 to surface complete data.
• -9999 is NPN's null sentinel. Tools translate this to undefined so you never see it in responses.
• Species_id=3 (red maple) returns empty. Known upstream NPN bug. The tool handles it gracefully (empty arrays, not errors).

Example prompts

After configuring, try:

• What phenology has been recorded near Port Angeles, WA over the last five years?
• Find all NPN stations in Arizona that observe saguaro cactus.
• Look up the NPN species ID for "mayapple".
• When do magnolias typically bloom in Washington state?
• Which states have the most NPN stations?
• Get all 2024 phenometric records for Acer rubrum (red maple) in WA.

Notes for AI agents

• For "what's blooming near me?" call get_active_phenology_nearby first. It composes the multi-step lookup (WKT, stations, site-level data) so you don't have to.
• For "when does X bloom in region Y?" use get_site_level_data with state filter. Returns mean first/last yes dates per phenophase.
• Avoid calling get_observations without species filtering. Single station-year results can be 80+ MB.
• All tools return the same Result<T>-style envelope. Success returns { ok: true, data }; error returns isError: true with structured { ok: false, error: { source, message, statusCode? } }.

License

MIT. See the root LICENSE.