OceanMCP

A monorepo of independently installable MCP servers for ocean and coastal data workflows.

Servers

| Server | PyPI | Description | |--------|------|-------------| | coops-mcp | | NOAA CO-OPS tides, water levels, currents, meteorological data | | erddap-mcp | | Universal ERDDAP data access across 80+ public servers | | nhc-mcp | | NHC storm tracks, advisories, HURDAT2 best track data | | recon-mcp | | Hurricane reconnaissance data (HDOB, Vortex Data Messages, ATCF fixes) | | stofs-mcp | | NOAA STOFS storm surge forecasts and observation validation | | ofs-mcp | | NOAA OFS regional ocean model forecasts (water level, temperature, salinity) | | rtofs-mcp | | NOAA RTOFS global ocean forecasts (SST, salinity, currents, SSH) via HYCOM THREDDS | | ww3-mcp | | GFS-Wave (WAVEWATCH III) forecasts and NDBC buoy wave observations | | ndbc-mcp | | NOAA NDBC buoy observations — waves, SST, wind, pressure (1,300+ stations) | | winds-mcp | | NWS surface wind observations from ASOS / AWOS stations | | usgs-mcp | | USGS Water Services — streamflow, NWS/NWPS flood status, peak events | | goes-mcp | | NOAA GOES-18/19 satellite imagery — visible, infrared, water vapor | | adcirc-mcp | | ADCIRC model setup debugging, parameter lookup, config validation | | schism-mcp | | SCHISM model setup debugging, parameter lookup, config validation |

No API keys required for the data servers — all servers above use free, publicly available datasets. The exceptions are hpc-system-mcp and ufs-runner-mcp (below), which require actual NOAA RDHPCS cluster access and, in ufs-runner-mcp's case, submit real Slurm jobs.

Additional servers live in this repo but are not yet published to PyPI (install from source): vdatum-mcp (vertical datum conversion), hpc-system-mcp and ufs-runner-mcp (NOAA RDHPCS / UFS-Coastal HPC workflows — require cluster access), and alert-mcp (CO-OPS threshold alerting).

There is also a 19th server in this repo, nos-workflow-mcp, which is experimental and still a work in progress (its full implementation lives on an unmerged branch — main carries only a minimal subset). It's intentionally left out of the table above and not yet ready for general use.

Quick Start

Install from PyPI

# uvx (recommended) — runs without permanent install, like npx for Python
uvx coops-mcp

# pip — install into current environment
pip install coops-mcp

# pipx — install in isolated environment with CLI entry point
pipx install coops-mcp

Replace coops-mcp with any published server: erddap-mcp, nhc-mcp, recon-mcp, stofs-mcp, ofs-mcp, rtofs-mcp, ww3-mcp, ndbc-mcp, winds-mcp, usgs-mcp, goes-mcp, adcirc-mcp, schism-mcp.

Install from source

git clone https://github.com/oceanmodeling/ocean-mcp.git
cd ocean-mcp/servers/coops-mcp  # or any other server directory under servers/
uv sync

Configure your MCP client

Add to your MCP settings (e.g., project .mcp.json):

Using PyPI packages (recommended):

{
  "mcpServers": {
    "coops": {
      "command": "uvx",
      "args": ["coops-mcp"]
    },
    "erddap": {
      "command": "uvx",
      "args": ["erddap-mcp"]
    },
    "nhc": {
      "command": "uvx",
      "args": ["nhc-mcp"]
    },
    "recon": {
      "command": "uvx",
      "args": ["recon-mcp"]
    },
    "stofs": {
      "command": "uvx",
      "args": ["stofs-mcp"]
    },
    "ofs": {
      "command": "uvx",
      "args": ["ofs-mcp"]
    },
    "rtofs": {
      "command": "uvx",
      "args": ["rtofs-mcp"]
    },
    "ww3": {
      "command": "uvx",
      "args": ["ww3-mcp"]
    }
  }
}

<details> <summary>Using local source checkout</summary>

{
  "mcpServers": {
    "coops": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/ocean-mcp/servers/coops-mcp", "python", "-m", "coops_mcp"]
    },
    "erddap": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/ocean-mcp/servers/erddap-mcp", "python", "-m", "erddap_mcp"]
    },
    "nhc": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/ocean-mcp/servers/nhc-mcp", "python", "-m", "nhc_mcp"]
    },
    "recon": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/ocean-mcp/servers/recon-mcp", "python", "-m", "recon_mcp"]
    },
    "stofs": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/ocean-mcp/servers/stofs-mcp", "python", "-m", "stofs_mcp"]
    },
    "ofs": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/ocean-mcp/servers/ofs-mcp", "python", "-m", "ofs_mcp"]
    },
    "rtofs": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/ocean-mcp/servers/rtofs-mcp", "python", "-m", "rtofs_mcp"]
    },
    "ww3": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/ocean-mcp/servers/ww3-mcp", "python", "-m", "ww3_mcp"]
    }
  }
}

</details>

Example Queries

With servers configured, you can ask your AI assistant naturally:

CO-OPS queries:

• "Get current water levels at The Battery, NY"
• "Find tide stations near Miami Beach"
• "What are the flood statistics for Charleston, SC?"
• "Show me the sea level trend at San Francisco"

ERDDAP queries:

• "Search for sea surface temperature datasets on CoastWatch"
• "Get chlorophyll data off the California coast for January 2024"
• "What ERDDAP servers cover the US East Coast?"
• "List all glider datasets on IOOS Gliders ERDDAP"

NHC queries:

• "Are there any active tropical cyclones right now?"
• "Show me Hurricane Katrina's track"
• "Search for Category 5 hurricanes in the Atlantic"
• "What is the forecast track for the active storm?"

Recon queries:

• "List HDOB reconnaissance flights for the 2024 Atlantic season"
• "Get HDOB flight-level observations from October 7, 2024"
• "Show me the Vortex Data Messages for Hurricane Milton"
• "Get ATCF aircraft fix data for storm AL14 2024"

STOFS queries:

• "Get the STOFS water level forecast for The Battery, NY"
• "Compare STOFS forecast vs observations at Boston for the past 24 hours"
• "What are the top stations with highest predicted water levels?"
• "Find STOFS stations within 50 km of New Orleans"

OFS queries:

• "What OFS models cover the Chesapeake Bay?"
• "Get the water level forecast at lat 38.98, lon -76.48 from CBOFS"
• "Compare CBOFS water level with CO-OPS observations at station 8571892"
• "List available NGOFS2 forecast cycles for today"

RTOFS queries:

• "Get the RTOFS sea surface temperature forecast at 40°N, 74°W"
• "Show me a temperature depth profile in the Gulf Stream"
• "What RTOFS datasets are currently available on HYCOM THREDDS?"
• "Get a vertical transect of salinity from Miami to Bermuda"
• "Show the SST area forecast for the Gulf of Mexico"

WW3 queries:

• "What GFS-Wave model grids are available?"
• "Find NDBC buoys near Cape Hatteras"
• "Get the latest wave observations from buoy 41025"
• "Show me the GFS-Wave forecast at 35°N, 75°W for the next 48 hours"
• "Compare the wave forecast against buoy 41025 observations"

Cross-server queries:

• "Find CO-OPS stations near this ERDDAP buoy location"
• "Compare tide station data with nearby ERDDAP satellite SST"
• "Is there an active hurricane threatening the Gulf Coast? If so, show me the STOFS surge forecast for New Orleans"
• "Get the CBOFS water level forecast at Chesapeake Bay and compare it with the CO-OPS tide prediction for the same station"
• "Show me sea surface temperature from ERDDAP near The Battery and the current water level from CO-OPS"
• "Which NHC storms have hit the Gulf Coast historically? Show the CO-OPS flood statistics for impacted stations"
• "Get the STOFS gridded forecast and the OFS water level forecast at the same location and compare them"

Architecture

Each server is fully self-contained with its own pyproject.toml, dependencies, and tests. See docs/architecture.md for details on shared conventions.

Shared patterns across servers:

• FastMCP for MCP server framework
• httpx for async HTTP clients
• Pydantic models, with Literal[...] types for constrained parameters
• Read-only tools by default — the data-fetching servers only fetch, never modify, anything. ufs-runner-mcp (submits/cancels Slurm jobs) and alert-mcp (creates/pauses/deletes in-memory alert state) are the two exceptions, each with a small number of tools marked readOnlyHint=False
• response_format output as markdown (default), json, geojson, or image, depending on the tool
• Actionable error messages with suggestions

See CONVENTIONS.md for the full response/behavior contract (output caps, retrieved_at, error format, HTTP retries, dependency bounds).

Citation

If you use this project in your research or work, please cite:

@software{jisan2025oceanmcp,
  author    = {Jisan, Mansur Ali},
  title     = {Ocean MCP: Real-Time Marine Data, MCP-Native},
  year      = {2025},
  url       = {https://github.com/oceanmodeling/ocean-mcp},
  note      = {MCP servers for NOAA CO-OPS, ERDDAP, NHC, Recon, STOFS, OFS, RTOFS, WW3, NDBC, USGS, winds, GOES, ADCIRC, and SCHISM data}
}

License

Licensed under the MIT License. You may use, modify, and distribute this software under the MIT terms. See LICENSE for details.