Question about MCP server tools: How does client infer what the tool's parameter values should be?

[bsloan666]

Pre-submission Checklist

• I have checked that this question would not be more appropriate as an issue in a specific repository
• I have searched existing discussions and documentation for answers

Question Category

• Protocol Specification
• SDK Usage
• Server Implementation
• General Implementation
• Documentation
• Other

Your Question

In the MCP server python example, weather.py (https://modelcontextprotocol.io/docs/develop/build-server), tools are defined using the @mcp.tool() decorator on global methods get_alerts(state) and get_forecast(lat, long). Does the caller of the tool (the client?) have any more information than the function signature of the tool's method and perhaps the parameter names? Is there some magic by which the mcp library can introspect the tool method's docstrings or the method's code?

Otherwise, I would expect that get_alerts(state: str) would not provide the client enough of a clue as to what a valid input param should be -- in this case a 2-letter USA state code. Are detailed docstrings a requirement. Are there other recommended ways to provide parameter hints to the client?

Also, this screen-shot from the tutorial is confusing (apologies if this is the wrong forum):

"View result from get-weather from weather (local)" I assume get_forecast() may have been named get_weather() at some point.

Thanks in advance!

[grzetich]

You bring up some good points. Let's address them one-by-one:

Are detailed docstrings a requirement?

No. As written in the example, the LLM will probably read the docstring for the get_alerts async function and see that state is expected to be a two-character US state abbreviation. However, that's not a durable pattern for production since it relies on model compliance with no runtime validation.

Are there other recommended ways to provide parameter hints to the client?

There's no officially recommended way in the MCP docs, but you can add validation using Pydantic's Field (which is already a transitive dependency of the MCP SDK).

async def get_alerts(
    state: Annotated[str, Field(min_length=2, max_length=2, description="Two-letter US state code (e.g. CA, NY)")]
) -> str:
    """Get weather alerts for a US state."""

This generates a proper JSON Schema with constraints:

{
  "name": "get_alerts",
  "description": "Get weather alerts for a US state.",
  "inputSchema": {
    "type": "object",
    "properties": {
      "state": {
        "type": "string",
        "minLength": 2,
        "maxLength": 2,
        "description": "Two-letter US state code (e.g. CA, NY)"
      }
    },
    "required": ["state"]
  }
}

As for the screenshot, it looks like it may be the result of using the get_weather tool as described in the specification.