Contributing

[jurasofish]

Hey! Awesome project, I am literally in the middle of building something pretty much the same as this - but this is better. I'd love to take my efforts and incorporate them into this project. Are you open to contributions?

Some things I'd love to add immediately

• strict mypy
• Clear docs on logging - since the MCP server runs in the background it's nice to log to a file you can tail, having a nice recipe for this in the docs would be great
• Ability to have the server log every request and response so you can watch things as they happen
• (Handle complex inputs #31) Ensure use of x: Annotated[int, Field(description="...")] results in the description in the json schema (looks like it already does but just needs some tests)
• (Handle complex inputs #31) Handle complex inputs like x: list[int], x: list[str] | str | None, x: MyModel - this is actually hard because at least with claude desktop it seems unable to output anything other than an object mapping from string to string/float/number - so lists/nulls/sub-models end up output as json strings and thus fail validation. I have code that does this that I can contribute, I can see how it would fit into FastMCP without much trouble, just need to rework the use of validate_call
• (Handle complex inputs #31) Handle things like x: Annotated[int, Field(1)]\n y: int so like required arg after one with a "default" via pydantic's Field.
• (Handle complex inputs #31) Remove context items from the generated JSON schema

less critical things

• docs on how to test tools/resources/etc with FastMCP
• middleware, to inject things into requests/responses. For example, I might like to inject into every response how many characters are in it. Or I might like to transform certain inputs before they're handled for all tools. Possibly overkill
• maybe dependency injection. Handy for testing (better typing than overriding env vars), handy in cases where various args need to validate things against some common state. Possibly overkill

[jlowin]

@jurasofish that all sounds awesome! Would welcome contributions.

I actually didn’t look closely enough to realize that the input spec was so limited and would love to expand it. ControlFlow may have some helpful code, though it doesn’t have the issue of non-structured inputs.