AimYear
Measured conclusion: Deep-tested. L2 outcome: tools_ok.
Accountability partner in your agent: one yearly aim, one main step a day, a timer your agent runs.
| Author | io.github.Yahhi |
|---|---|
| Access | remote |
| Publisher key required | No |
| Measured state | Deep-tested |
| L1 liveness | alive |
| L1 latency | 159 ms |
| Last measured | 2026-10-24 |
Add to your agent
Copy this into your MCP client config:
{
"mcpServers": {
"AimYear": {
"url": "https://mcp.aimyear.com/mcp",
"transport": "streamable-http"
}
}
}
Measured tools (51)
get_goalread-onlyThe one yearly goal: kind (count or done), metric, starting value, current value, target, deadline, progress, levers (hard/comfortable) and other projects. Call first in most flows.
Hints: read-only yes · destructive no · idempotent yes · open-world no
check_goal_draftread-onlyCheck a goal that is not saved yet against the rubric (a countable metric with value today and target, or a done-or-not result you can check yes or no; deadline within 12 months; pace per week). Changes nothing. Returns problems[] and next_question: loop until problems is empty, then call set_goal. Prefer count; use done only when no honest number exists.
Parameters:
kind— string — count (default): a number from baseline to target, e.g. "15 paying customers". done: a one-time result checked yes or no, e.g. "launch v2 in the stores"; no metric, baseline or target, progress is the share of stages donetitle— stringmetric— string — the one number and its unit, e.g. "paying customers"target— numberbaseline— number — value todaydeadline— string — YYYY-MM-DD in the user's timezone
Hints: read-only yes · destructive no · idempotent yes · open-world no
set_goaldestructiveSave the yearly goal, or replace it with the next one. Rejects with problems[] if the rubric fails. Keeps levers and projects. area is which of the five life areas the goal belongs to.
Parameters:
title(required) — stringdeadline(required) — string — YYYY-MM-DD in the user's timezonewhy— stringarea— string — default moneykind— string — count (default): a number from baseline to target, e.g. "15 paying customers". done: a one-time result checked yes or no, e.g. "launch v2 in the stores"; no metric, baseline or target, progress is the share of stages donemetric— string — required for kind counttarget— number — required for kind countbaseline— number — value today; required for kind count
Hints: read-only no · destructive yes · idempotent yes · open-world no
record_progressRecord the current value of the goal metric ("I now have 4 customers"), or of a countable milestone when milestoneId is given ("I have talked to 7 people"). For a done-or-not goal, 1 marks it done. One value per date: calling again for the same date corrects it. Says when the target is reached; a countable milestone that reaches its target is done.
Parameters:
value(required) — numberdate— string — YYYY-MM-DD in the user's timezonemilestoneId— string — a countable milestone instead of the goal
Hints: read-only no · destructive no · idempotent yes · open-world no
set_leversdestructiveReplace the list of levers: the recurring kinds of goal work. At least one must be kind=hard (the work the user avoids). keywords/repos are hints for classifying sessions from their task text. Pass existing ids to keep them.
Parameters:
levers(required) — array
Hints: read-only no · destructive yes · idempotent yes · open-world no
set_projectsdestructiveReplace the list of other projects (paid work, other products, life routines), so their sessions count as other work or life, not goal work. Pass the whole list: leave one out to remove it, change its name or hints to edit it. Pass existing ids to keep them.
Parameters:
projects(required) — array
Hints: read-only no · destructive yes · idempotent yes · open-world no
get_pathread-onlyMilestones with kind, progress (recorded value or steps done), pace (ahead/on_pace/behind) and their steps. current_milestone_id is the one to work on.
Hints: read-only yes · destructive no · idempotent yes · open-world no
set_milestonesdestructiveReplace the list of 2–6 ordered milestones under the goal, each countable (metric and target) or measured by its steps, with a due date before the goal deadline. Pass each existing id, or that milestone's steps are deleted. To change one milestone, use edit_milestone.
Parameters:
milestones(required) — array
Hints: read-only no · destructive yes · idempotent yes · open-world no
edit_milestoneChange one milestone: title, kind (count or steps), metric, target, due date or status. status=retired drops it without finishing; done or retired moves the path to the next milestone. Its steps are kept.
Parameters:
milestoneId(required) — stringdue— string — YYYY-MM-DD in the user's timezonekind— string — count: progress is a recorded value against target ("talk to 20 people"); steps: progress is steps done ("implement the onboarding flow"), no metric needed. Default: count when metric and target are given, else steps; an existing milestone keeps its kindtitle— stringmetric— stringstatus— stringtarget— number
Hints: read-only no · destructive no · idempotent yes · open-world no
add_stepsAdd concrete steps to a milestone. A step is an action that fits a few work blocks ("email 10 people from the waitlist"), not "do marketing". Abstract steps are rejected. Put a hard-lever step first; preparation only where it blocks it (warns otherwise).
Parameters:
milestoneId(required) — stringsteps(required) — arrayop_id— string — for apps replaying queued requests: an id the app minted before its first attempt and reuses on retries, so the write is applied once and a retry gets the first answer back. Do not invent one; an agent calling interactively leaves it out
Hints: read-only no · destructive no · idempotent no · open-world no
edit_stepChange one step: title, status (todo/doing/done), order, lever, estimate.
Parameters:
stepId(required) — stringorder— integertitle— stringstatus— stringleverId— stringestimateMin— integer
Hints: read-only no · destructive no · idempotent yes · open-world no
delete_stepdestructiveDelete one step from the path.
Parameters:
stepId(required) — string
Hints: read-only no · destructive yes · idempotent no · open-world no
next_stepsread-onlyOpen steps of the current milestone, hard-lever steps first: the candidates for today's main step.
Parameters:
count— integer
Hints: read-only yes · destructive no · idempotent yes · open-world no
get_commitmentsread-onlyRecurring weekly time: commitments (school runs, training, classes), overhead (sleep, meals) and focus windows.
Hints: read-only yes · destructive no · idempotent yes · open-world no
set_commitmentsdestructiveReplace the list of recurring weekly commitments. days are ISO weekdays 1=Mon..7=Sun; use timesPerWeek instead when there are no fixed days. Pass existing ids to keep them.
Parameters:
commitments(required) — array
Hints: read-only no · destructive yes · idempotent yes · open-world no
get_capacityread-onlyWeekly time budget out of 168 hours: overhead, hours committed per area, focus windows, and what is still unaccounted for. Nothing is assumed.
Hints: read-only yes · destructive no · idempotent yes · open-world no
get_planning_contextread-onlyEverything needed to plan a day: goal and pace, next_steps, yesterday's summary and undone items, week balance, days since a hard lever. Defaults to tomorrow. The agent plans; then call set_plan.
Parameters:
date— string — YYYY-MM-DD in the user's timezone
Hints: read-only yes · destructive no · idempotent yes · open-world no
get_planread-onlyThe plan for a day (default today).
Parameters:
date— string — YYYY-MM-DD in the user's timezone
Hints: read-only yes · destructive no · idempotent yes · open-world no
set_plandestructiveReplace the plan for a day. Exactly one item must have isMainStep=true (the step toward the goal), ideally linked by stepId from next_steps and on a hard lever. Warns when no item is on a hard lever. Other items: paid work, family, workout, etc. with their area. Items that stay keep their done state.
Parameters:
items(required) — arraydate— string — YYYY-MM-DD in the user's timezone
Hints: read-only no · destructive yes · idempotent yes · open-world no
edit_plan_itemChange one plan item: mark it done or not done, rename it, change its estimate, area or links, or make it the main step. done=true also completes its linked path step. For a weekly routine, count says how many units this tick did (talked to 2 people today). For a routine, level minimum says only its bad-day minimum was done: it counts for the day, reported apart from full.
Parameters:
itemId(required) — stringarea— stringdate— string — YYYY-MM-DD in the user's timezonedone— booleancount— integer — units done in this tick, for a routine with perWeek; above 0 ticks it, 0 unticks itlevel— string — routines only; sets done unless done is giventitle— stringstepId— stringleverId— stringprojectId— stringisMainStep— booleanestimateMin— integer
Hints: read-only no · destructive no · idempotent yes · open-world no
get_routinesread-onlyRoutines: small steps that repeat on fixed weekdays, e.g. "talk with one person on Reddit", or that count up to a number a week, e.g. "talk to 5 people a week", each with an optional bad-day minimum. The server adds them to each matching day's plan, so every device and agent sees the same checks. With no routines yet, returns ready-made starters for the aim's area to offer as they are.
Hints: read-only yes · destructive no · idempotent yes · open-world no
set_routinesdestructiveReplace the list of routines: small steps that repeat on fixed weekdays. From today on, each one appears as a plan item on its days; tick it with edit_plan_item, skip it for one day with delete_plan_item. set_plan keeps them. days are ISO weekdays 1=Mon..7=Sun, default every day. perWeek makes it a weekly count: each tick adds its count (default 1), and once the week reaches perWeek it leaves the plan until Monday. Link it to the goal with leverId. minimum is the small version for a bad day ("read one thread" for "talk to one person"). Pass existing ids to keep them.
Parameters:
routines(required) — array
Hints: read-only no · destructive yes · idempotent yes · open-world no
delete_plan_itemdestructiveRemove one item from a day's plan. A routine item is skipped for that day only.
Parameters:
itemId(required) — stringdate— string — YYYY-MM-DD in the user's timezone
Hints: read-only no · destructive yes · idempotent no · open-world no
set_day_modeTracking for a day: full (everything recorded, the default) or off (timer and reminders only, nothing stored, no plan required).
Parameters:
mode(required) — stringdate— string — YYYY-MM-DD in the user's timezone
Hints: read-only no · destructive no · idempotent yes · open-world no
get_nowread-onlyWhat is happening now: running sessions with attention shares, the timer, today's main step, today's minutes, the streak and a one-line nudge.
Hints: read-only yes · destructive no · idempotent yes · open-world no
start_sessionStart a session of work and, if none is running, the timer. Returns a verdict: ok | stop | plan_first. On "stop" NO session is created: relay the message, ask the user; if they insist, call again with override=true. Several sessions may run at once (two terminals) and share attention by weight. Apps replaying a queued start pass op_id, their own session_id and start_at; an agent leaves all three out.
Parameters:
task(required) — string — what you are about to do, specificarea— string — default workop_id— string — for apps replaying queued requests: an id the app minted before its first attempt and reuses on retries, so the write is applied once and a retry gets the first answer back. Do not invent one; an agent calling interactively leaves it outitemId— string — plan item idstepId— string — path step idweight— number — attention weight vs other sessions at the same time, default 1leverId— stringminutes— integer — length for a down timer, or the stand-up interval for an up timeroverride— boolean — true only after the user confirmed past a stopstart_at— string — when the work began as ISO 8601 with offset, e.g. 2026-10-20T14:05:00+02:00. For apps delivering a request queued offline; an agent calling interactively leaves it out. Rejected before the session start; a clock more than 2 minutes ahead of the server is taken as nowdirection— string — up (default) counts elapsed time and runs until stopped, with a stand-up reminder every workMin; down is a fixed timebox, for a breakexclusive— boolean — true: nothing else was happening thenprojectId— stringworkplace— stringsession_id— string — for apps: the session id the app minted itself. Create-if-absent: if a session with this id exists it is returned and nothing is created. An agent leaves it out; the server mints the id
Hints: read-only no · destructive no · idempotent no · open-world no
add_noteAdd a one-line note to a running session (this connection's, or session_id). May classify a session that was unclassified.
Parameters:
note(required) — stringop_id— string — for apps replaying queued requests: an id the app minted before its first attempt and reuses on retries, so the write is applied once and a retry gets the first answer back. Do not invent one; an agent calling interactively leaves it outsession_id— string — from get_now; required with op_id
Hints: read-only no · destructive no · idempotent no · open-world no
split_attentionSet attention weights of the running sessions at once, e.g. {"abc": 0.7, "def": 0.3}. Applies to the whole session. For a finished session use edit_session.
Parameters:
shares(required) — object
Hints: read-only no · destructive no · idempotent yes · open-world no
stop_sessionStop a running session with a summary and outcome. Pass session_id from get_now; without it, only this connection's single running session is stopped, never another one. The timer keeps running for other sessions. outcome=done completes the linked plan item or path step. Without end_at or op_id it refuses a session that already ended: use edit_session for that. With end_at (apps delivering a stop queued offline) a second stop on the same session is merged, whatever order they arrive in: the earliest end wins, done wins over any other outcome, the first summary stays and a different later one goes to the notes; the answer says applied: ended | shortened | recorded.
Parameters:
summary(required) — stringoutcome(required) — stringop_id— string — for apps replaying queued requests: an id the app minted before its first attempt and reuses on retries, so the write is applied once and a retry gets the first answer back. Do not invent one; an agent calling interactively leaves it outend_at— string — when the work ended as ISO 8601 with offset, e.g. 2026-10-20T14:05:00+02:00. For apps delivering a request queued offline; an agent calling interactively leaves it out. Rejected before the session start; a clock more than 2 minutes ahead of the server is taken as nowsession_id— string — from get_now; required with op_id or end_atattention_share— number — corrects the share in hindsight
Hints: read-only no · destructive no · idempotent no · open-world no
stop_timerdestructiveStop the timer ("pause", "stop", "stand up") and every running session with it. There is no paused state: resuming starts a new session. Pass summary and outcome to save them on the sessions it closes, or call stop_session first. Returns the break length.
Parameters:
outcome— string — default partialsummary— string
Hints: read-only no · destructive yes · idempotent no · open-world no
add_sessionAdd past time that was not timed: minutes (ending now, or at end), or start+end as HH:MM on date. For calls, children, workout, paid work, or work done away from the agent. Apps replaying a queued entry pass op_id, so a retry does not count the time twice.
Parameters:
task(required) — stringend— string — HH:MM local timearea— string — default workdate— string — YYYY-MM-DD in the user's timezoneop_id— string — for apps replaying queued requests: an id the app minted before its first attempt and reuses on retries, so the write is applied once and a retry gets the first answer back. Do not invent one; an agent calling interactively leaves it outstart— string — HH:MM local timeitemId— string — plan item idstepId— string — path step idweight— number — attention weight vs other sessions at the same time, default 1leverId— stringminutes— integerexclusive— boolean — true: nothing else was happening thenprojectId— string
Hints: read-only no · destructive no · idempotent no · open-world no
edit_sessionFix a session, running or finished: start and end (HH:MM on date; an end before the start means it ran past midnight), task, area, project, lever, step, plan item, summary, outcome, attention share, weight. It is classified again after a change. Use it when a timer was forgotten or a session was labelled wrong.
Parameters:
session_id(required) — stringend— string — HH:MM local timearea— string — default workdate— string — local date of start and end; default the session's own datetask— stringstart— string — HH:MM local timeitemId— string — plan item idstepId— string — path step idweight— number — attention weight vs other sessions at the same time, default 1leverId— stringoutcome— stringsummary— stringexclusive— boolean — true: nothing else was happening thenprojectId— stringattention_share— number — share of the session's wall time that was real attention
Hints: read-only no · destructive no · idempotent yes · open-world no
delete_sessiondestructiveDelete a session, running or finished, so it no longer counts. Pass session_id from get_now or list_sessions; without it, only this connection's single running session is deleted, never another one.
Parameters:
op_id— string — for apps replaying queued requests: an id the app minted before its first attempt and reuses on retries, so the write is applied once and a retry gets the first answer back. Do not invent one; an agent calling interactively leaves it outsession_id— string — required with op_id
Hints: read-only no · destructive yes · idempotent no · open-world no
list_sessionsread-onlySessions in a date range (default the last 7 days), with ids, optionally filtered by area or project.
Parameters:
to— string — YYYY-MM-DD in the user's timezonearea— stringfrom— string — YYYY-MM-DD in the user's timezoneprojectId— string
Hints: read-only yes · destructive no · idempotent yes · open-world no
find_gapsread-onlyUntimed stretches of 20+ minutes between the first and last session of a day. Ask the user what happened there.
Parameters:
date— string — YYYY-MM-DD in the user's timezone
Hints: read-only yes · destructive no · idempotent yes · open-world no
summarize_dayread-onlyFull summary of a day: wall and attention minutes per area, classification and lever, sessions, gaps, overlaps, overrides.
Parameters:
date— string — YYYY-MM-DD in the user's timezone
Hints: read-only yes · destructive no · idempotent yes · open-world no
summarize_weekread-onlyA week, Monday to Sunday (to today for the current week): main-step days, goal share, hard share, area balance, overrides, streak, milestone pace.
Parameters:
date— string — any day of the week, default today
Hints: read-only yes · destructive no · idempotent yes · open-world no
get_insightsread-onlyWeek (the last 7 days) or month (the last 30 days) ending on to (default today): minutes per day on the aim, other work and life; hard vs comfortable share of aim time; an hour × weekday heatmap of aim minutes (6:00–23:00, Mon–Sun) with its hard part and the peak; main-step days; and comebacks (active days after 2+ days away, over 90 days). Wall minutes of finished sessions, as the app shows them.
Parameters:
to— string — last day, default todayrange— string — default week
Hints: read-only yes · destructive no · idempotent yes · open-world no
find_leaksread-onlyWhere time leaks: comfortable goal work while the main step waited, overrides, unplanned work, gaps, switching, missed main steps, days without a plan. Default the last 7 days.
Parameters:
to— string — YYYY-MM-DD in the user's timezonefrom— string — YYYY-MM-DD in the user's timezone
Hints: read-only yes · destructive no · idempotent yes · open-world no
get_settingsread-onlyTimezone, where reminders go (desktop or mobile), timer settings, and which networks get progress posts (post_to).
Hints: read-only yes · destructive no · idempotent yes · open-world no
set_settingsChange any of: timezone (IANA name, e.g. Europe/Madrid; it sets where each day starts and ends), workplace for reminders, stand-up interval workMin, breakMin, longBreakEvery, longBreakMin, standup on/off, post_to (networks for progress posts).
Parameters:
post_to— array — which connected networks get the day's progress post (the app sets it); [] for nonestandup— booleanworkMin— integerbreakMin— integertimezone— stringworkplace— stringlongBreakMin— integerlongBreakEvery— integer
Hints: read-only no · destructive no · idempotent yes · open-world no
add_deviceRegister a device token so server reminders (stand up, break over) can reach it. The phone app registers itself; use this for testing or a manual token.
Parameters:
token(required) — stringplatform(required) — stringlabel— string
Hints: read-only no · destructive no · idempotent yes · open-world no
list_devicesread-onlyWhich devices can receive reminders, and when each was last seen.
Hints: read-only yes · destructive no · idempotent yes · open-world no
delete_devicedestructiveRemove a device so it stops receiving reminders.
Parameters:
token(required) — string
Hints: read-only no · destructive yes · idempotent no · open-world no
list_agentsread-onlyAgents connected to this AimYear account (Claude, ChatGPT, Claude Code, the AimYear app…): name, client_id, when it connected and when it was last active. current marks the one calling.
Hints: read-only yes · destructive no · idempotent yes · open-world no
delete_agentdestructiveDisconnect one agent: its sign-in for this account is deleted and it can no longer read or change anything until it connects again. Only on the user's explicit request. client_id from list_agents.
Parameters:
client_id(required) — string
Hints: read-only no · destructive yes · idempotent no · open-world no
add_social_accountConnect WIP (wip.co) or Bluesky so AimYear can post the day's progress there when the user says Post (Pro). WIP: api_key from wip.co settings. Bluesky: handle and an app password (Settings → Privacy and security → App passwords), never the account password. The login is checked, stored encrypted and never shown again. The app's connect screen usually calls this; if the user pastes a key into the chat, use it once and do not repeat it.
Parameters:
network(required) — stringhandle— string — Bluesky only, e.g. name.bsky.socialapi_key— string — WIP onlyapp_password— string — Bluesky only: xxxx-xxxx-xxxx-xxxx
Hints: read-only no · destructive no · idempotent yes · open-world yes
list_social_accountsread-onlyConnected networks (handle, since when), which of them get progress posts, and whether the user has Pro. Never returns any key or password.
Hints: read-only yes · destructive no · idempotent yes · open-world no
delete_social_accountdestructiveDisconnect a network and delete its saved login. Posts already made stay there.
Parameters:
network(required) — string
Hints: read-only no · destructive yes · idempotent no · open-world no
get_post_draftread-onlyThe day's progress post drafted from the record (aim and progress, main step done or not, minutes on the aim and on hard work, routines, comebacks), at most 300 characters, ending with aimyear.com; and per network whether it is ready, already posted, or skipped because the user already posted there that day. Posts nothing. Pro.
Parameters:
date— string — YYYY-MM-DD in the user's timezone
Hints: read-only yes · destructive no · idempotent yes · open-world yes
add_postPost the day's progress, only after the user explicitly said to post. text is the draft as the user edited it (default: the draft from get_post_draft). Goes to the networks chosen in settings, or the ones given. A network where the user already posted that day themselves is skipped. Each day goes to each network once: a retry never posts twice. Today or yesterday only. Pro.
Parameters:
date— string — YYYY-MM-DD in the user's timezonetext— stringnetworks— array — default: the networks chosen in settings
Hints: read-only no · destructive no · idempotent yes · open-world yes
Measured by MCPYP probing. Untested fields are omitted, never estimated. Listing is always free; MCPYP never proxies calls or stores user tokens.