ERDDAP MCP Server

Access oceanographic and environmental data from ERDDAP servers worldwide through Claude Desktop viatwo complete MCP implementations: local stdio and remote HTTP.

🚀 Now with dynamic server loading fromerddaps.json- 63+ ERDDAP servers available!

This repository providesboth local and remote MCP server implementations:

📍 Local MCP Server (erddapy_mcp_server.py)

- Traditional stdio-based MCP serverfor local Claude Desktop use
- 4 comprehensive ERDDAP toolsfor data discovery and access
- Easy setupvia claude_desktop_config.json
- No network dependencies- runs completely locally

☁️ Remote MCP Server (erddap_remote_mcp_oauth.py)

- HTTP-based MCP serverfor cloud deployment
- Production-readywith fly.io deployment configuration
- mcp-remote proxy compatiblefor Claude Desktop integration
- Same 4 core toolsoptimized for remote performance

🚨 CRITICAL: Remote MCP Connection Requirements

Claude Desktop does NOT support direct remote MCP connections!You MUST use themcp-remoteproxy for remote servers.

Claude Desktop (stdio) ↔ mcp-remote proxy ↔ Remote MCP Server (HTTP)

ERDDAP (Environmental Research Division's Data Access Program) is a data server that provides simple, consistent access to scientific datasets in common file formats. These MCP servers make ERDDAP's powerful oceanographic data accessible to AI assistants through natural language queries.

ERDDAP servers may experience severe issues when accessed through VPNs(NordVPN, ExpressVPN, etc.), including:

- 404 errorson valid datasets
- Truncated or incomplete datadownloads
- Connection timeoutsand failed requests

Solution:Disable your VPN before using these tools. ERDDAP servers often block or rate-limit VPN traffic, causing unreliable behavior.

Option 1: Local MCP Server (Recommended for Getting Started)

2. Configure Claude Desktop:Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{ "mcpServers": { "erddap-local": { "command": "python", "args": ["/path/to/erddapy_mcp_server.py"] } } }

Windows Users:Use%APPDATA%\Claude\claude_desktop_config.jsonand double backslashes in paths:

{ "mcpServers": { "erddap-local": { "command": "python", "args": ["C:\\Users\\YourName\\path\\to\\erddapy_mcp_server.py"] } } }

3. Restart Claude Desktop:After updating the configuration, restart Claude Desktop. The server will start automatically and the ERDDAP tools will be available.

Option 2: Remote MCP Server (Access the Cloud Instance)

{ "mcpServers": { "erddap-remote": { "command": "npx", "args": ["mcp-remote", "https://erddap2mcp.fly.dev/"] } } }

3. Restart Claude Desktop:The remote server will be accessible through the mcp-remote proxy.

You:"Show me available ERDDAP servers"

Claude:Here are some well-known ERDDAP servers you can access:

- NOAA CoastWatch(https://coastwatch.pfeg.noaa.gov/erddap) - Satellite and in-situ oceanographic data
- IOOS ERDDAP(https://erddap.ioos.us/erddap) - Integrated Ocean Observing System data
- Marine Institute Ireland(https://erddap.marine.ie/erddap) - Irish marine data
- Ocean Networks Canada(https://data.oceannetworks.ca/erddap) - Canadian ocean data
- GCOOS ERDDAP(https://gcoos5.geos.tamu.edu/erddap) - Gulf of Mexico data

# Deploy from the erddap2mcp directory fly deploy

app = 'erddap2mcp' primary_region = 'mia' [http_service] internal_port = 8000 force_https = true auto_stop_machines = 'stop' auto_start_machines = true [[vm]] memory = '1gb' cpu_kind = 'shared' cpus = 1

- ✅Automatic HTTPS- SSL certificates managed automatically
- ✅Auto-scaling- Machines start/stop based on traffic
- ✅Global CDN- Fast access worldwide
- ✅Zero-downtime deploys- Seamless updates

# Build container docker build -t erddap-mcp-server . # Run locally docker run -p 8000:8000 erddap-mcp-server # Deploy to other platforms: # - AWS Lambda (requires HTTPS setup) # - Railway/Render (usually provide HTTPS automatically) # - Google Cloud Run # - Azure Container Instances

After configuring Claude Desktop, restart it and check if the ERDDAP tools appear in the tools list.

# Test the server directly (for debugging) python erddapy_mcp_server.py # Send test commands: echo '{"jsonrpc": "2.0", "method": "tools/list", "id": 1}' | python erddapy_mcp_server.py

# Test basic connectivity curl http://localhost:8000/ # Test MCP protocol curl -X POST http://localhost:8000/ \ -H "Content-Type: application/json" \ -d '{"jsonrpc": "2.0", "method": "tools/list", "id": 1}' # Test with mcp-remote proxy npx mcp-remote http://localhost:8000/ --test

You can run both local and remote servers simultaneously:

{ "mcpServers": { "erddap-local": { "command": "python", "args": ["/Users/rdc/src/mcp/erddap2mcp/erddapy_mcp_server.py"] }, "erddap-remote": { "command": "npx", "args": ["mcp-remote", "https://erddap2mcp.fly.dev/"] } } }

This gives you the full local toolset plus cloud accessibility!

- server_url: ERDDAP server URL (defaults to NOAA CoastWatch)
- protocol: Either "tabledap" (tabular data) or "griddap" (gridded data)
- dataset_id: The dataset identifier
- variables: List of variables to retrieve
- constraints: Dictionary of constraints (e.g., time/space bounds)
- Start with local: Local server has easier setup and no proxy requirement
- Use remote for sharing: Remote server can be accessed by multiple users
- Check metadata first: Useget_dataset_infobefore downloading data
- Use constraints: Limit data requests to avoid timeouts
- Choose correct protocol: tabledap for tabular, griddap for gridded data

- "No such file": Check Python path in configuration
- "Connection failed": Verify server is running
- Tool errors: Check stderr for debug output

- "No tools available": Ensure mcp-remote proxy is installed (npm install -g mcp-remote)
- "Connection failed": Verify server URL is accessible and uses HTTPS
- Protocol errors: Check server logs withfly logs -a erddap2mcp

- Timeouts: Reduce spatial/temporal extent of requests
- Protocol errors: Specify correct protocol (tabledap vs griddap)
- Server unavailable: Some ERDDAP servers may be temporarily down

This remote implementation represents months of debugging the Remote MCP mystery:
- Direct SSE connections- Claude Desktop doesn't support this
- Config file remote URLs- Only works for local stdio servers
- Connector UI attempts- Also doesn't support direct connections

Themcp-remoteproxy requirement was buried in third-party documentation. This critical piece enables Claude Desktop to talk to remote MCP servers via HTTP.

- ERDDAPwas developed by Bob Simons at NOAA's Environmental Research Division. Learn more at theERDDAP website.
- erddapyis the official Python client for ERDDAP, developed byFilipe Fernandesand the IOOS community. Visit theerddapy documentation.
- mcp-remoteproxy enables remote MCP connections to Claude Desktop

{ "name": "Your ERDDAP Server", "short_name": "YES", "url": "https://your-erddap.org/erddap/", "public": true }

Access the Protein Data Bank (PDB) for 3D structures of proteins and nucleic acids, with tools for structural analysis and comparison.