Skip to content

Installing AQUAVIEW MCP

The AQUAVIEW MCP server is hosted at:

https://mcp.aquaview.org/mcp

It speaks HTTP transport (the modern MCP standard, formerly called Streamable HTTP). No installation, no API keys, no rate-limit signup. Just point your MCP client at the URL.

This page covers every supported client. Pick yours.

Table of contents

Claude Code (CLI)

One command:

claude mcp add --transport http aquaview https://mcp.aquaview.org/mcp

Restart your session and ask Claude: "List the AQUAVIEW collections."

Claude Desktop

  1. Open Claude → Settings → Developer → Edit Config.
  2. Merge the snippet below into the mcpServers block:
{
  "mcpServers": {
    "aquaview": {
      "type": "http",
      "url": "https://mcp.aquaview.org/mcp"
    }
  }
}
  1. Quit and relaunch Claude Desktop. You should see AQUAVIEW's four tools in the tool picker (the small hammer icon under the input box).

Older Claude Desktop builds without HTTP transport support can use the mcp-remote shim. See Troubleshooting.

Anthropic API direct

The Anthropic API supports remote MCP servers as a first-class parameter — your application doesn't need to host or proxy anything.

from anthropic import Anthropic

client = Anthropic()

response = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=4096,
    mcp_servers=[{
        "type": "url",
        "url": "https://mcp.aquaview.org/mcp",
        "name": "aquaview",
    }],
    messages=[{
        "role": "user",
        "content": "Find Argo float profiles within 200 km of Hawaii in March 2026.",
    }],
)

print(response.content)

TypeScript (@anthropic-ai/sdk) takes the same mcp_servers array.

ChatGPT Custom Connector

ChatGPT Plus/Pro/Team workspaces with Custom Connectors enabled can add AQUAVIEW directly:

  1. Open ChatGPT → Settings → Connectors → Add custom.
  2. Name: AQUAVIEW. Server URL: https://mcp.aquaview.org/mcp. Transport: Streamable HTTP.
  3. Save. The connector appears in the model picker; toggle it on for any chat.

OpenAI Agents SDK

The OpenAI Agents SDK supports MCP through the hosted-MCP tool type. No proxy, no shim:

from agents import Agent, Runner
from agents.mcp.server import MCPServerStreamableHttp

aquaview = MCPServerStreamableHttp(
    name="aquaview",
    params={"url": "https://mcp.aquaview.org/mcp"},
)

agent = Agent(
    name="Ocean Researcher",
    model="gpt-5",
    instructions="You answer ocean and atmospheric science questions using AQUAVIEW.",
    mcp_servers=[aquaview],
)

result = Runner.run_sync(agent, "Show me Hurricane Ian buoy data.")
print(result.final_output)

The TypeScript SDK exposes the same MCPServerStreamableHttp class.

Google Gemini Agent SDK

The Gemini API exposes MCP servers as native tools through the google-genai Python SDK:

from google import genai
from google.genai import types

client = genai.Client()

aquaview = types.Tool(
    mcp=types.McpToolConfig(
        server_url="https://mcp.aquaview.org/mcp",
        transport="http",
    )
)

response = client.models.generate_content(
    model="gemini-2.5-pro",
    contents="What's the current sea surface temperature near the Florida Keys?",
    config=types.GenerateContentConfig(tools=[aquaview]),
)

print(response.text)

Cursor

  1. Settings → MCP → Add new MCP Server.
  2. Name: aquaview. Type: http. URL: https://mcp.aquaview.org/mcp.
  3. Cursor reconnects automatically. The status dot turns green when the four tools are loaded.

Cline

In your VS Code workspace, edit .cline/mcp_settings.json:

{
  "mcpServers": {
    "aquaview": {
      "type": "streamable-http",
      "url": "https://mcp.aquaview.org/mcp"
    }
  }
}

Continue

Edit ~/.continue/config.json:

{
  "mcpServers": [
    {
      "name": "aquaview",
      "transport": {
        "type": "http",
        "url": "https://mcp.aquaview.org/mcp"
      }
    }
  ]
}

Zed

Zed reads MCP servers from ~/.config/zed/settings.json:

{
  "context_servers": {
    "aquaview": {
      "command": {
        "transport": "http",
        "url": "https://mcp.aquaview.org/mcp"
      }
    }
  }
}

Verifying it works

After connecting in any client, run this prompt:

List the AQUAVIEW collections and tell me how many there are.

You should see the model call list_collections and report 68 collections. If it reports a different number or fails to call the tool, see Troubleshooting.

Troubleshooting

"No tools found" / connector shows as offline. Confirm the URL is exactly https://mcp.aquaview.org/mcp — note the trailing /mcp path component is required.

Client only supports stdio (older Claude Desktop, some IDE plugins). Use the mcp-remote bridge:

{
  "mcpServers": {
    "aquaview": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.aquaview.org/mcp"]
    }
  }
}

Corporate proxy / firewall. The endpoint requires outbound HTTPS to mcp.aquaview.org on port 443. The connection is a long-lived streaming HTTP request; some intercepting proxies break it. If you see frequent disconnects, add mcp.aquaview.org to your bypass list.

Still stuck? Open an issue with your client name, version, and any error logs.