class

Smith::MCP::ServerHandle

Inherits Reference < Object

The lifecycle of one server: start it, keep it, restart it once if it dies, then give up on it.

The give-up is deliberate. A server that crashes twice is broken, and retrying it on every call would turn every turn into a restart loop — slower than having no tool at all, and far harder to read from a transcript.

Constructors

new(spec : ServerSpec, name : String, timeout : Time::Span = Client::DEFAULT_TIMEOUT, startup_timeout : Time::Span = Client::STARTUP_TIMEOUT)

Source

Instance methods

call(tool : String, arguments : JSON::Any) : ToolResult

Source

error

Source

lost?

Source

name

The sanitised server name, as it appears in mcp__<server>__<tool>.

Source

on_lost

Wired by whoever registered the tools, so they can be withdrawn when the server is given up on. The manager knows the registry; this file does not.

Source

on_lost=(on_lost : Proc(ServerHandle, Nil) | Nil)

Wired by whoever registered the tools, so they can be withdrawn when the server is given up on. The manager knows the registry; this file does not.

Source

running?

Source

spec

Source

start

False rather than raising: one server that will not start must never take the session with it.

Source

stderr_tail

The server's own stderr, kept so a failed handshake can say what the process actually complained about.

Source

stop

Source

tools

Source