Documentation
¶
Overview ¶
Package mcp implements the toolkit MCP server. It exposes the same read-only category surface as the headless `toolkit get` CLI as a set of MCP tools that an agent (Claude Code, Claude Desktop, any MCP-aware client) can call directly over stdio — no shell out, no output parsing.
The handlers reuse the existing loader composite (internal/infra/loader), so any improvement to data loading (partial-tolerance, variable defaults, etc.) is shared between CLI and MCP automatically.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type Server ¶
type Server struct {
// contains filtered or unexported fields
}
Server is the toolkit's MCP server. Build with NewServer, then call Run.
func NewServer ¶
NewServer constructs a server that exposes read-only category tools. cfg supplies the startup env defaults; each tool call may override env_type / env_region / env_realm per-call.
func (*Server) Run ¶
Run blocks until any of:
- stdin reaches EOF (the MCP client closed the pipe),
- ctx is canceled,
- the underlying transport returns a fatal error.
Returns nil on a clean client disconnect or ctx cancel, otherwise the transport's error. Uses stdio: stdin reads JSON-RPC frames, stdout writes them. Callers must keep stdout free of any other output (the toolkit logger writes to a file by default — see internal/cli/mcp.go).
Single-shot: a second Run on the same *Server reuses the SDK's session list rather than reinitializing cleanly. Construct a fresh Server (via NewServer) if you need to restart.
Source Files