Skip to content

Daily Agent MCP — OpenClaw Skill

What this is

dailyagent is the user's personal productivity database, exposed as an MCP server. It is the single source of truth for:

  • Tasks (Franklin Covey priority A1-C9, with rollover, spaces, goals, recurrence)
  • Habits and habit logs
  • Journal entries (with mood 1-5 and full-text search)
  • Workouts (templates, logs, per-exercise sets/reps/weight/type)
  • Focus sessions (Pomodoro-style, optionally linked to a task)
  • Goals (with progress, categories, target dates)
  • Spaces (projects/areas that group tasks/goals/habits)
  • Weekly reviews, daily briefings, and insight cache (read + write)

The server runs on the user's VPS behind Tailscale. Reads and writes go straight to Postgres. You do not store anything locally — always go through the MCP tools.

Connection

The MCP server is already registered in the user's OpenClaw config as dailyagent. You do not need to configure auth — it's handled by the gateway (bearer token in the OpenClaw config). If tool calls fail with 401 / connection errors, stop and tell the user — do not fall back to local files or markdown templates.

Core rules

  1. Always use MCP tools. Never invent data, never fall back to scratch notes or file-based tracking. If a tool doesn't exist for what the user wants, say so.
  2. Today's date is authoritative. When the user says "today," use the current date (YYYY-MM-DD). The server also defaults most date fields to today if omitted.
  3. Tasks roll over. list_tasks with no date returns today's tasks plus incomplete tasks from previous days. You don't need to manually pull overdue tasks.
  4. Priority format is A1-C9. Letter (A critical / B important / C nice-to-have) + digit 1-9 for sub-ordering. If the user just says "high priority" → use A1. If they say "low" → C1. Default if unspecified → B1 (the tool handles this — just omit).
  5. Dates are always YYYY-MM-DD. Parse natural language ("tomorrow", "next Monday") into that format before calling tools.
  6. Writes are authoritative. Don't ask "should I save this?" — if the user said "log my workout" / "add a task" / "journal this," just do it.

Tools

Tasks

  • list_tasks(date?, space_id?) — today's + overdue if no date; else that specific date
  • create_task(title, notes?, priority?, task_date?, space_id?, goal_id?) — priority A1-C9, defaults to B1
  • update_task(task_id, title?, notes?, priority?, task_date?, done?)
  • complete_task(task_id)
  • delete_task(task_id)

Habits

  • list_habits(include_archived?)
  • get_habit_stats(habit_id, days?) — default 30 days
  • create_habit(name, description?, frequency?, target_days?, color?) — frequency daily|weekly; target_days ISO 1-7 (Mon=1, Sun=7)
  • toggle_habit(habit_id, date?) — idempotent toggle; defaults to today

Journal

  • get_journal_entries(date?|from?,to?, limit?) — specific day or a range
  • search_journal(query) — substring search
  • create_journal_entry(content, entry_date?, mood?) — mood 1-5; upserts if entry for that date exists

Workouts

  • list_workout_logs(date?|from?,to?) — most recent 20 if no filter
  • list_workout_templates()
  • log_workout(name, log_date, duration_minutes?, notes?, exercises?) — exercises is a JSON string array; each entry: {name, type?: "strength"|"timed"|"cardio", sets?, reps?, weight?, duration_seconds?, notes?}

Focus

  • get_focus_sessions(from?, to?)
  • get_focus_stats() — today's totals
  • start_focus_session(duration_minutes, task_id?, break_minutes?) — duration 1-480, break 0-120
  • complete_focus_session(session_id)

Goals

  • list_goals(status?) — status active|completed|abandoned
  • create_goal(title, description?, category?, target_date?) — category health|career|personal|financial|learning|relationships|other
  • update_goal(goal_id, title?, description?, status?, progress?) — progress 0-100
  • log_goal_progress(goal_id, progress) — just the progress percentage

Spaces (projects)

  • list_spaces()
  • create_space(name, description?)
  • update_space(space_id, name?, description?, status?) — status active|paused|completed

Weekly Reviews

  • get_weekly_review(week_start?) — latest if omitted
  • save_weekly_review(week_start, content) — markdown

Daily Briefings

  • get_daily_briefing() — today's, if saved
  • save_daily_briefing(briefing_date?, content) — markdown; defaults to today

Insights (AI-generated insight cards)

  • get_insights() — today's cached insights
  • save_insights(cache_date?, insights) — insights must be an array or object

Calendar / summary

  • get_day_summary(date) — tasks + habits + journal + workouts + focus for that day
  • get_week_summary(week_start) — aggregate across a week

Prompt templates (via MCP prompts)

The server exposes prompt templates you can load for structured outputs. Use these when the user asks for something generative:

  • daily_planning — plan today
  • morning_briefing — the daily 7am-style briefing
  • end_of_day_review — end-of-day reflection
  • weekly_review — the weekly review structure
  • weekly_trends — trend analysis over a week
  • productivity_report — stats + narrative
  • habit_analysis — habit consistency deep-dive
  • goal_check_in — per-goal progress check-in
  • goal_planning — set up a new goal
  • space_planning — plan a project/space
  • week_planning — plan out the upcoming week
  • journal_prompt — journaling starter for today
  • workout_suggestion — suggest a workout based on recent logs

Pattern: load the prompt → fill it with fresh data (call the read tools) → generate → save the result with the matching save tool (e.g. save_daily_briefing, save_weekly_review, save_insights).

What a loaded prompt looks like

When you call a prompt, the server runs the read queries itself and returns a messages array of {role: "user", content: {type: "text", text: "..."}} entries. You don't need to call list_tasks / list_habits beforehand — the prompt has already pulled fresh data and embedded it as JSON inside the text. Just generate a response to the returned messages, then call the matching save tool.

Example — loading morning_briefing returns a single user message whose text looks roughly like:

Good morning! Give me a quick briefing for 2026-04-20.

## What's on My Plate
- 7 tasks pending (2 are A-priority)
- 4 habits to track today
- 3 active goals in progress

### Task Details
[{"id":"...","title":"Ship MCP prompt docs","priority":"A1","done":false,"task_date":"2026-04-20"}, ...]

### Habits
[{"id":"...","name":"Morning walk","completed_today":false}, ...]

Keep it brief: 3-4 bullet points on what matters most today, then a one-sentence motivational close.

So the flow is: prompt_load("morning_briefing") → you generate the briefing text → save_daily_briefing({content: "..."}) → deliver. Prompts that take args (productivity_report, weekly_review, goal_planning, space_planning) accept them as the prompt arguments, not as a separate tool call.

Prompts with args: - productivity_report(from, to) — both YYYY-MM-DD, required - weekly_review(week_start?) — defaults to this week's Monday - goal_planning(goal_id) — resolve the ID via list_goals first - space_planning(space_id?) — omit for cross-space planning

All other prompts take no args.

Resources (read-only URIs)

For quick contextual reads you can fetch resources instead of calling tools:

  • dailyagent://dashboard — today at a glance
  • dailyagent://tasks/today, dailyagent://tasks/overdue
  • dailyagent://habits/today, dailyagent://habits/streaks
  • dailyagent://journal/today, dailyagent://journal/recent
  • dailyagent://workouts/recent
  • dailyagent://focus/today
  • dailyagent://goals/active
  • dailyagent://spaces/list
  • dailyagent://briefing/today
  • dailyagent://calendar/today, dailyagent://calendar/week
  • dailyagent://review/latest

Common patterns

"What's on my list today?" → list_tasks with no args. Show them, grouped by priority letter.

"Add a task: X" → create_task({title: "X"}). Don't ask for priority — the default B1 is fine unless the user specifies.

"Mark X done" / "I finished X" → list_tasks → find by title → complete_task(task_id).

"Log my workout" → Ask for what was done if not stated, then log_workout with the exercises array. Include type on each exercise (strength/timed/cardio).

"How am I doing on habits?" → list_habits → for each, get_habit_stats. Summarize completion rates.

"Journal for today: ..." → create_journal_entry({content: "..."}). If they also mention a mood number, include it.

"Start a 25 minute focus session on X" → list_tasks → find X → start_focus_session({duration_minutes: 25, task_id}).

Morning briefing (7am cron) → Load morning_briefing prompt → pull dashboard resource → generate → save_daily_briefing → deliver to Telegram.

Weekly review (Sunday evening cron) → Load weekly_review prompt → pull calendar/week + get_week_summary → generate → save_weekly_review.

What NOT to do

  • Do not create parallel tracking in markdown files, local notes, or templates. The MCP server is the only store.
  • Do not skip the save step after generating a briefing or review — always persist it via the save tool so the dashboard shows it.
  • Do not retry failed writes indefinitely. If a tool returns an error, surface it to the user with the error text.
  • Do not guess IDs. Always resolve a task/habit/goal by listing first, then act on the returned id.