Docs exports: JSON llms.txt llms-full.txt Agent Files

module Shards::Docs::MCPLifecycle

Overview

MCP Server Lifecycle Management

The shards mcp command manages the runtime lifecycle of MCP servers distributed via .mcp-shards.json. This completes the pipeline from distribution (handled by shards install) to execution.

Commands

shards mcp                         # Show server status (default)
shards mcp start [server_name]     # Start all or one server
shards mcp stop [server_name]      # Stop all or one server
shards mcp restart [server_name]   # Restart all or one server
shards mcp logs <name> [--no-follow] [--lines=N]

Runtime state

All managed state lives in .shards/mcp/:

servers.json: PID, port, timestamps per server
<name>.log: per-server stdout/stderr logs
bin/: cached builds for crystal_main servers

Process management

Servers are spawned via Process.new (non-blocking) with output redirected to log files. PID tracking uses LibC.kill(pid, 0). Shutdown sends SIGTERM, waits 5 seconds, then SIGKILL if needed. Stale PIDs are detected and cleaned on every status check.

Name resolution

Server names use the existing namespacing from .mcp-shards.json (e.g., my_shard/explorer). Partial name matching is supported: explorer finds my_shard/explorer if unambiguous.

See MCPManager, Commands::MCP.

Defined in:

docs.cr

Ashard