Proposal: Allow any return type in tool functions

[cpunion]

Issue Description

Currently, tool functions can only return String directly or implement the IntoCallToolResult trait for custom types. This limits the flexibility and requires unnecessary boilerplate.

Suggested Enhancement

Allow tool functions to return any serializable type by:

1. Making #[tool(param)] attribute optional on function parameters
2. Automatically converting return values to CallToolResult using serde serialization

Benefits

• Easier function definitions
• Cleaner test code
• More intuitive API
• Less boilerplate implementation
• Better type safety

Example (Current vs Proposed)

Current approach:

#[tool(description = "Calculate sum")]
fn sum(&self, #[tool(param)] a: i32, #[tool(param)] b: i32) -> String {
    (a + b).to_string()
}

Proposed approach:

#[tool(description = "Calculate sum")]
fn sum(&self, a: i32, b: i32) -> i32 {
    a + b
}

This enhancement would make the API more ergonomic while maintaining all current functionality.

[cpunion]

Comparison with Python SDK

The Python SDK already supports this pattern with a cleaner API:

@mcp.tool()
def add(a: int, b: int) -> int:
    """Add two numbers"""
    return a + b

async def main():
    print(add(1, 2))  # Normal function call: 3
    print(await mcp.call_tool("add", {"a": 1, "b": 2}))  # Via MCP: 3

if __name__ == "__main__":
    asyncio.run(main())

This demonstrates how returning native types directly makes the API more intuitive.

[4t145]

We can add a auto serialization wrapper to tell sdk using a JSON serialization policy, it could be looks like this.

#[tool(description = "Calculate sum")]
fn sum(&self, a: i32, b: i32) -> Json<i32> {
    Json(a + b)
}

[cpunion]

@4t145 Thanks! The Json wrapper approach looks great.

[4t145]

Supported in #78

[cpunion]

Great! Thank.