activeing123/mcptoon v0.8.39

v0.8.39 - flag ownership across 20 commands, --yaml removed, --format fixed

on GitHub

[0.8.39] - 2026-11-12

Changed

• Flag ownership now also covers the action-first commands — plugin, kb,
  session, resources, prompts, complete (issue #41, final batch). Their flags
  follow a subcommand (mcptoon plugin install <dir> --force), so each set is the
  union over the actions, read from _cmd_plugin_* and the command's own module
  (kb.py, session_log.py, mcp_extras.py). A flag standing where the action
  belongs is still the command's own error, which says more than ours would
  (Unknown plugin subcommand: --force. Available: scan | install | list | remove).
  The registry now covers 20 commands; auth, bake, openapi and graphql
  stay out because the pre-pass hands them the rest of the line verbatim and their
  flags belong to a sub-module.
• Flag ownership now covers call, install, activity, sync, serve,
  skills and retrieve (issue #41, second batch). Each set was read from the
  source that parses it — the handler in cli.py, or the command's own module
  (skills.py, serve.py) — and cross-checked against the invocations the shipped
  docs show. auth/bake/openapi/graphql stay out on purpose: the pre-pass
  hands them the rest of the line verbatim, so their flags belong to a sub-module and
  cannot be enumerated here. Two consequences of the wider set: --help now always
  passes (ten commands carry their own help page, where it is not consumed by the
  pre-pass and stays in the argument list), and a new test sweeps the docs and fails
  if a registered command would reject a flag those docs promise — so the table can
  keep growing one command at a time without documenting a rejected invocation. The
  error's "did you mean" hint prints names now, not a Python list repr.
• --format now says who owns it, and validates its value. It is manifest's
  export selector (openai|openapi|mcp|json|human), but the global pre-pass consumes
  it for every command — it has to, because mcptoon --format mcp manifest puts the
  flag before the command. Measured on 0.8.38: list --format json,
  list --format toon and list --format bogus all printed the same
  940-character human list and exited 0 (fmt is set by --json/--toon/--slim,
  never by --format). A command with no export now exits 2 and points at the flags
  that do set the format, and manifest --format mpc (a typo) exits 1 with the
  allowed list instead of silently dumping JSON — which is what export_manifest's
  documented fallback used to produce. manifest/tools keep working, in either
  position.
• A flag a command does not own is now a usage error. mcptoon list --verbose
  printed the human server list and exited 0: --verbose is a real flag, just
  manifest's, so the global pre-pass tolerated it and list silently dropped it —
  a correctly spelled wrong flag was quieter than the typo list --verbos
  (exit 2). list, inspect, search and manifest now declare the flags they own
  explicitly and reject anything else with exit 2, naming the command that does own
  the flag ('--verbose' is an option of: manifest). A command joins that registry
  only once its flags are enumerated by hand — never inferred from a handler's
  docstring, which would reject flags the code supports but the prose omits. Every
  other command keeps the previous behaviour until its turn (issue #41).
  MCPTOON_STRICT_FLAGS=0 still downgrades this and the unknown-option error to a
  warning.

Fixed

• --yaml is no longer accepted. It was registered as a pass-through for a
  per-command YAML renderer that was never written: ArgParserFactory declared the
  flag, no code read the parsed attribute, and output.render has no yaml branch —
  so mcptoon list --yaml printed human text and exited 0 while list --json
  printed data, and mcptoon --help never mentioned the flag at all (found
  2026-11-12 while auditing the option-ownership surface, #49). It is now an unknown
  option — exit 2 — in either position, and the factory baseline no longer offers it.
  YAML output will come back per command, with a renderer and tests, when that topic
  is taken up.