GribStream Weather MCP

GribStream Weather MCP is a hosted Model Context Protocol server for AI access to weather forecast data from GribStream, including NOAA and ECMWF model output.

It helps AI agents discover forecast datasets such as NOAA GFS, HRRR, and NBM or ECMWF IFS, resolve exact forecast variables and levels, validate request bodies, build runnable GribStream Weather API requests, and execute live read-only weather queries after OAuth authorization.

Official GribStream weather API homepage:

https://gribstream.com/

Hosted MCP endpoint:

https://gribstream.com/mcp

Use Cases

Use GribStream MCP when you want an AI tool to:

• List public GribStream weather datasets.
• Choose a NOAA, ECMWF, or other forecast model for a region, horizon, or weather task.
• Query GFS, HRRR, NBM, IFS, and other numerical weather prediction model data exposed by GribStream.
• Find exact variable selectors, levels, and aliases.
• Build a /timeseries request for values by valid time.
• Build a /runs request to compare forecasts across model runs.
• Validate a GribStream API request before running it.
• Execute live /timeseries or /runs queries after OAuth authorization.
• Retrieve dense forecast grids as downloadable files without embedding the complete result in the conversation.
• Generate a copy-pasteable curl command for the GribStream Weather API.

The hosted MCP server exposes discovery, selector lookup, request building, and validation tools before OAuth. Live /timeseries and /runs query tools require OAuth. Tool descriptions begin with either NO AUTH / PUBLIC / READ-ONLY or AUTH REQUIRED / DATA QUERY so AI clients can route metadata tasks and live data-query tasks correctly. When a client connects through OAuth, GribStream asks the user to sign in and select an active GribStream API token. MCP access tokens are scoped to that selected API token, and the raw API token is not shown to the MCP client.

Connect

Use the hosted endpoint exactly as shown in any MCP client that supports remote Streamable HTTP:

https://gribstream.com/mcp

Do not include trailing punctuation in the URL.

Generic configuration shape:

{
  "mcpServers": {
    "gribstream": {
      "url": "https://gribstream.com/mcp"
    }
  }
}

LobeHub quick import configuration:

{
  "mcpServers": {
    "gribstream": {
      "type": "http",
      "url": "https://gribstream.com/mcp"
    }
  }
}

Some clients require an explicit transport field:

{
  "mcpServers": {
    "gribstream": {
      "type": "streamable-http",
      "url": "https://gribstream.com/mcp"
    }
  }
}

OAuth

OAuth is needed for live query tools. Most MCP clients should discover OAuth from the hosted endpoint automatically, either during connector setup or when the client requests authenticated tools. If a client asks for manual OAuth 2.0 fields, use:

Authorization URL: https://gribstream.com/authorize
Token URL: https://gribstream.com/token
Client ID: gribstream-mcp-public
Client Secret: leave empty; do not enter a value
Scopes: leave empty
Resource or audience, if requested: https://gribstream.com/mcp

OAuth discovery endpoints:

https://gribstream.com/.well-known/oauth-protected-resource/mcp
https://gribstream.com/.well-known/oauth-authorization-server

The OAuth client is public and uses PKCE with token_endpoint_auth_method set to none. Native desktop clients may use loopback redirect URIs such as http://localhost:{port}/..., http://127.0.0.1:{port}/..., or http://[::1]:{port}/.... A setup form that requires a non-empty client secret is asking for a confidential-client flow and is not compatible with this public MCP OAuth flow.

Server Features

LobeHub and other MCP crawlers may call MCP tools "skills". The live GribStream MCP server exposes the following skills/tools, prompts, and resources.

Skill List

Public read-only skills/tools are visible before OAuth. OAuth live query skills/tools may be hidden from tools/list until the client completes OAuth or sends a valid Authorization: Bearer token.

Prompt List

Resource List

Live Query Output

Authenticated live query tools support CSV, JSON, NDJSON, and Parquet results. Small CSV, JSON, and NDJSON results can be returned inline. Set delivery=url for dense grids or large results to receive a short-lived signed HTTPS resource_link to the complete file instead.

• URL delivery works with CSV, JSON, NDJSON, and Parquet. Parquet always uses URL delivery.
• Download the exact signed URL once without an API token. This consumes the MCP result; it is not a separate direct GribStream API request.
• The 10 MB max_bytes limit applies only to inline MCP results. URL artifacts use a separate server-configured size limit, so omit max_bytes when using delivery=url.
• For dense grids, prefer Parquet when a compatible reader is already installed; otherwise use CSV with delivery=url. Do not install a dependency or write a Parquet parser solely for one result.
• Use one grid request and only split after an actual request or artifact-size error. If splitting is necessary, split time first, variables second, and space last.
• structuredContent includes metadata such as content_type, rows, bytes, columns, schema, delivery, cache_hit, signed_url_expires_at, and suggested_filename when applicable.
• Inline tabular results also include a preview plus request_hash, response_hash, and result_hash. Use suggested_filename when saving multiple results so repeated requests or comparisons do not overwrite earlier files.
• Rows are streamed from parallel extraction and are not guaranteed to be sorted. Sort by forecasted_time for /timeseries, and by forecasted_at plus forecasted_time for /runs, before plotting or comparing time series.

Links

• Official GribStream weather API homepage: https://gribstream.com/
• Endpoint: https://gribstream.com/mcp
• MCP page: https://gribstream.com/mcp
• Icon: https://gribstream.com/gribstream_small.png
• OpenAPI: https://gribstream.com/docs/openapi.yaml
• AI setup docs: https://gribstream.com/ai
• API dashboard: https://gribstream.com/app/dashboard

Try It

After connecting GribStream MCP to an AI tool, ask:

Use GribStream MCP to build a GFS 2m temperature timeseries request for Houston for tomorrow at 12:00 UTC. Validate the request and return a runnable curl command.

If your client has completed OAuth, you can also ask it to run the query:

Use GribStream MCP to run a GFS 2m temperature timeseries query for Houston for tomorrow at 12:00 UTC and return CSV.

For a dense-grid visualization:

Using GribStream, plot the latest available ICON-D2 2 metre temperature forecast over Germany at 0.025 degree resolution as a beautiful PNG.

Or: