Can AI replace Outstand?

Build me a self-hosted unified posting API to replace Outstand. Requirements:

- Node + Fastify + better-sqlite3, one headless service on localhost:8080.
  No web UI: my own apps are the only consumer, every key in .env.
- One normalized post shape in zod (text, media, targets, scheduledAt,
  idempotencyKey) and one adapter per network mapping it onto that
  platform's payload. Adapters export capabilities, publish, and metrics;
  nothing platform-specific leaks into the core.
- A capability table per adapter (character limit, media count, mime types,
  aspect ratios, max video seconds). POST /posts returns 422 naming which
  target rejected what, before a single network call goes out.
- Ship the three networks that need no app review: Bluesky (@atproto/api),
  Mastodon, Telegram. Adding a fourth must mean writing one adapter file
  and touching nothing else.
- A token-bucket limiter per network seeded with their published limits so
  calls queue instead of returning 429. POST /posts answers 202 with a job
  id and a worker drains the queue.
- Idempotency: replaying a POST with the same idempotencyKey returns the
  original job instead of posting twice, and every target gets its own
  delivery row so partial success is readable.
- GET /posts/:id/analytics folds each platform's metrics into one shape
  (impressions, likes, comments, shares) with null where a network does not
  report a field. Never invent a number to fill a gap.
- An MCP server over stdio exposing create_post, list_accounts, and
  get_analytics so Claude Code posts through it, plus an OpenAPI spec
  generated from the zod schemas at /openapi.json.
- Out of scope: Instagram, TikTok, YouTube, LinkedIn, Pinterest. Each needs
  my own developer app approved by the platform first, a review queue and
  not code. Say that in the README next to the three token setups.