Todoist Weekly Review MCP

An MCP server for doing a weekly review of your Todoist with Claude. It lists your overdue tasks with the signals that matter (days overdue, priority, how often you've postponed each one). Claude suggests a fix per task, you approve or veto each one in chat, and only the approved changes are applied. The server decides nothing on its own and makes no LLM calls.

v1 handles overdue tasks only. Roadmap and design history: PLAN.md, CONTEXT.md.

Quick start

npm install && npm run build
cp .env.example .env   # paste your token from Todoist → Settings → Integrations → Developer

Open Claude Code in this folder, approve the todoist-weekly-review server when asked, and say "run my weekly review".

Related MCP server: Todoist MCP Server Extended

Tools

Safety

• Nothing is written without your explicit per-item approval in chat.

• No delete exists. Tasks are retired by completing them or moving them to "Someday/Maybe" (created automatically if missing).

• Token lives in the gitignored .env only — never logged, never committed.

• Read errors are loud (bad token, rate limit, etc.) — never a silent empty list.

• A malformed change rejects the whole batch before anything is written.

Design notes

• "Times rescheduled" comes from the task's postponed_count field, so the plan-gated activity log isn't needed. Omitted when Todoist doesn't return it.

• Priority is inverted in the API: 4 = highest (the UI's P1). Passed through as-is.

• Endpoints verified against the live v1 OpenAPI spec. Two fixes vs. the draft plan: filtering is GET /tasks/filter?query=…, and moving a task is POST /tasks/{id}/move.

• Todoist quirk: a task due earlier today doesn't count as "overdue" until tomorrow.

• Days overdue are computed in your Todoist account's timezone.

Tests

npm test

28 unit tests (API mocked, no live calls) plus a smoke test that boots the real server over stdio without any token.