Struct McpCommand 

pub struct McpCommand {
    request: Request,
    reply: Sender<Result<Output, String>>,
    world: WorldId,
}

Available on crate feature mcp and non-WebAssembly only.

A request together with the channel its result is sent back on.

Fields

request: Request

reply: Sender<Result<Output, String>>

world: WorldId

The world a Request::Job runs against.

Implementations

impl McpCommand

pub fn new( job: impl FnOnce(&mut World) -> Result<Output, String> + Send + 'static, ) -> (Self, Receiver<Result<Output, String>>)

Build a command that runs a job, and the receiver its result will arrive on.

The command starts out aimed at the world the receiving thread owns directly; redirect it with with_world.

pub fn list_worlds() -> (Self, Receiver<Result<Output, String>>)

Build the command the list_worlds tool sends.

The host answers it with answer_worlds; the server holds no world list of its own.

fn build(request: Request) -> (Self, Receiver<Result<Output, String>>)

pub fn with_world(self, world: WorldId) -> Self

Direct the command at world.

pub const fn world(&self) -> WorldId

The world a job runs against.

pub const fn wants_worlds(&self) -> bool

Whether this is the list_worlds request, which the host answers with answer_worlds instead of dispatching a job.

pub fn answer_worlds(self, worlds: Vec<WorldEntry>)

Answer the list_worlds request with the worlds the host carries.

The host names every world a tool can reach — the host world itself and each world it routes commands to, including ones on other threads — each with the name it was spawned under.

pub fn fail(self, message: String)

Answer the command with an error without running it.

The host calls this when the world cannot be resolved, so the tool call still gets a readable message rather than a dropped channel.

pub fn dispatch(self, world: &mut World)

Run the command’s job against world and send its result back.

A list_worlds request has no job; the host is expected to answer it with answer_worlds after checking wants_worlds.

A job that panics is caught and answered as an error: a world usually shares its thread with others — the host’s own, or sibling sub-worlds — and an unwind out of a job would kill them all with it. The world the job died in may still be inconsistent; the error says so.

The send is best-effort: a tool call that has already been cancelled has no receiver left, and the result is simply dropped.