v0.1.0

Initial release — a Model Context Protocol server exposing the Missive REST API
to AI assistants in Claude Desktop and Claude Code.

Tools (36, all prefixed missive_)

• Contacts — list, get, create, update (with full infos/memberships).
• Contact books & groups — list contact books; list groups/organizations.
• Conversations — list (with required mailbox filter), get, update (close /
  reopen / assign / label / move / rename), merge, and list a conversation's
  messages / comments / drafts / posts.
• Messages — get (single or batched by id), find by RFC Message-ID, and
  create an incoming custom-channel message.
• Drafts — create a draft (email/SMS/custom-channel) with attachments,
  recipients, threading, labels, and routing. Saved for review — never sent.
• Posts — create an internal comment/note in a conversation.
• Shared labels — list, create, update.
• Teams — list, create, update.
• Users / Organizations — list.
• Responses (canned replies) — list, get, create, update.
• Tasks — list (with until cursor + filters), get, create, update.
• Self-documentation — missive_help with topics including usage, tools,
  and safety.

Safety model

• No external send. missive_create_draft saves a draft for a human to
  send; the send / send_at / auto_followup parameters are intentionally not
  exposed, enforced by a regression test.
• No delete tools.
• Internal posting and conversation merge are supported — they act only
  within Missive and never email anyone outside the team. Merge is marked
  irreversible.
• Tool annotations advertise read-only / idempotent / destructive intent so MCP
  clients can gate confirmation appropriately.

Authentication & configuration

• Static personal-access-token auth loaded from .env; the token never appears
  in any launcher config.
• Optional defaults to avoid repeating IDs: MISSIVE_DEFAULT_ORGANIZATION,
  MISSIVE_DEFAULT_CONTACT_BOOK, MISSIVE_DEFAULT_TEAM,
  MISSIVE_DEFAULT_FROM_ADDRESS (+ MISSIVE_DEFAULT_FROM_NAME), and
  MISSIVE_DEFAULT_ACCOUNT.
• npm run setup validates the token and prints your organizations, contact
  books, teams, and users, plus a ready-to-paste launcher config.

Architecture

• A hand-written fetch client with a concurrency cap, HTTP-429 retry honoring
  Retry-After, a per-request timeout distinguished from network errors, and
  uniform result shaping.
• MCP instructions surfaced to the client at connect time, so the assistant
  learns the API's rules and recipes up front.

Docs & tests

• README, a friendly step-by-step SETUP.md, a complete docs/TOOLS.md tool
  reference, CLAUDE.md, CONTRIBUTING.md, and SECURITY.md.
• A vitest suite covering the HTTP client (including the rate-limit semaphore and
  timeout handling), the helpers, the live MCP handshake, handler validation, and
  the no-send guarantee, plus doc-rot guards that keep the tool count in sync.
• GitHub Actions for CI and releases.