AimYear MCP Server
by yahhiio.github.Yahhi/aimyear-mcpv0.4.2
Accountability partner in your agent: one yearly aim, one main step a day, a timer your agent runs.
context tax
~6,445 tok/turn43 tools
security
92/100
cold start
hosted
freshness
Active1d ago
Our reading of AimYear, from the data
AimYear is an MCP server from yahhi, listed in the official MCP Registry.
Its 43 tools add about 6,445 tokens to every model turn (measured 2026-10-19).
Maintenance is active: the last public signal was in the last day (2026-10-18).
It runs as a hosted endpoint, so there is nothing to install.
Install AimYear MCP server
Install in Claude Code
claude mcp add --transport http aimyear https://mcp.aimyear.com/mcpInstall in Cursor
{
"mcpServers": {
"aimyear": {
"url": "https://mcp.aimyear.com/mcp"
}
}
}Install in Claude Desktop
{
"mcpServers": {
"aimyear": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://mcp.aimyear.com/mcp"
]
}
}
}Or add it as a custom connector under Settings → Connectors. The config above bridges via mcp-remote.
Install in VS Code
{
"servers": {
"aimyear": {
"type": "http",
"url": "https://mcp.aimyear.com/mcp"
}
}
}Install in Windsurf
{
"mcpServers": {
"aimyear": {
"serverUrl": "https://mcp.aimyear.com/mcp"
}
}
}Remote endpoints
- streamable-http
https://mcp.aimyear.com/mcp
Context tax
Connecting this server adds ~6,445 tokens of tool definitions to every model turn (heavy, 43 tools). Measured 2026-10-19 via its hosted endpoint; o200k tokenizer, worst case for clients that load all tools up front. How we measure
| Model | Per 1,000 turns | With prompt caching |
|---|---|---|
| Claude Opus 5.5 | $25.78 | $1.29 |
| Claude Sonnet 5.5 | $12.89 | $1.29 |
| Claude Haiku 5.5 | $0.644 | — |
Input prices as of 2026-10-06. Excludes your own prompt and output.
Tools (43)
edit_session395 tokFix 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.
set_commitments384 tokReplace 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.
start_session347 tokStart 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.
add_session318 tokAdd 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.
set_plan253 tokReplace 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.
set_routines240 tokReplace 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. Pass existing ids to keep them.
edit_plan_item215 tokChange 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.
set_goal188 tokSave 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.
set_milestones182 tokReplace the list of 2–6 ordered milestones under the goal, each with a metric, target and due date before the goal deadline. Pass each existing id, or that milestone's steps are deleted. To change one milestone, use edit_milestone.
set_levers176 tokReplace 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.
set_projects176 tokReplace 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.
set_settings175 tokChange 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.
check_goal_draft173 tokCheck a goal that is not saved yet against the rubric (countable metric, value today, target, deadline within 12 months, pace per week). Changes nothing. Returns problems[] and next_question: loop until problems is empty, then call set_goal.
add_steps172 tokAdd 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).
edit_milestone165 tokChange one milestone: title, 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.
list_sessions155 tokSessions in a date range (default the last 7 days), with ids, optionally filtered by area or project.
stop_session149 tokStop a running session (this connection's, or session_id) with a summary and outcome. The timer keeps running for other sessions. outcome=done completes the linked plan item or path step. Refuses a session that already ended: use edit_session for that.
edit_step147 tokChange one step: title, status (todo/doing/done), order, lever, estimate.
find_leaks144 tokWhere 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.
record_progress122 tokRecord the current value of the goal metric, e.g. "I now have 4 customers". One value per date: calling again for the same date corrects it. Says when the target is reached.
stop_timer120 tokStop 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.
set_day_mode116 tokTracking for a day: full (everything recorded, the default) or off (timer and reminders only, nothing stored, no plan required).
get_planning_context113 tokEverything 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.
add_device107 tokRegister 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.
split_attention106 tokSet 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.
summarize_week105 tokA week, Monday to Sunday (to today for the current week): main-step days, goal share, hard share, area balance, overrides, streak, milestone pace.
delete_plan_item100 tokRemove one item from a day's plan. A routine item is skipped for that day only.
summarize_day95 tokFull summary of a day: wall and attention minutes per area, classification and lever, sessions, gaps, overlaps, overrides.
find_gaps94 tokUntimed stretches of 20+ minutes between the first and last session of a day. Ask the user what happened there.
get_plan78 tokThe plan for a day (default today).
get_routines78 tokRoutines: small steps that repeat on fixed weekdays, e.g. "talk with one person on Reddit". The server adds them to each matching day's plan, so every device and agent sees the same checks.
add_note78 tokAdd a one-line note to a running session (this connection's, or session_id). May classify a session that was unclassified.
get_goal68 tokThe one yearly goal: metric, starting value, current value, target, deadline, levers (hard/comfortable) and other projects. Call first in most flows.
next_steps67 tokOpen steps of the current milestone, hard-lever steps first: the candidates for today's main step.
get_capacity66 tokWeekly time budget out of 168 hours: overhead, hours committed per area, focus windows, and what is still unaccounted for. Nothing is assumed.
delete_session65 tokDelete a session, running or finished, so it no longer counts. Without session_id it deletes this connection's running session.
get_now64 tokWhat is happening now: running sessions with attention shares, the timer, today's main step, today's minutes, the streak and a one-line nudge.
get_path62 tokMilestones with pace (ahead/on_pace/behind) and their steps. current_milestone_id is the one to work on.
get_commitments58 tokRecurring weekly time: commitments (school runs, training, classes), overhead (sleep, meals) and focus windows.
delete_step52 tokDelete one step from the path.
delete_device52 tokRemove a device so it stops receiving reminders.
get_settings48 tokTimezone, where reminders go (desktop or mobile), and timer settings.
list_devices47 tokWhich devices can receive reminders, and when each was last seen.
Security signals
92/100 — no known issues found as of 2026-10-19, from the checks below. This is not a guarantee of safety; review what a server can do before connecting it.
Deductions
- No license detected-5
- No security policy (SECURITY.md)-3
Checks
- ✓Repository hygiene (archived, activity, license, security policy)
- ○Package metadata (install scripts, provenance) — pending
- ○Tool descriptions (prompt-injection and hidden-text heuristics) — pending
- ○Source scan (secrets, CVEs, risky code) by vectra-guard — pending
Scan it yourself before installing with vectra-guard, the open-source agent security CLI. Scoring rules
Freshness
Active — last maintenance signal 1d ago. The newest of the signals below sets the band.
Last commit (default branch)
2026-10-18 · 1d ago · GitHub
Latest release
no data · GitHub
Package published
no data · npm/PyPI
Registry entry updated
2026-10-18 · 1d ago · official registry · v0.4.2
FAQ
›How do I install the AimYear MCP server in Claude Code?
Run: claude mcp add --transport http aimyear https://mcp.aimyear.com/mcp. For Cursor, VS Code, Claude Desktop and Windsurf, use the install tabs above.
›Does AimYear require an API key?
No required environment variables are declared in its published metadata.
›Can I use AimYear as a remote (hosted) MCP server?
Yes. It is a hosted MCP server; connect to its URL with any client that supports remote MCP.
›Is AimYear in the official MCP registry?
Yes, as io.github.Yahhi/aimyear-mcp.
›How many tokens does AimYear add to my context?
Its tool definitions total about 6,445 tokens, resent on every turn.
Sources and method
- Official MCP Registry entry
- updated 2026-10-18
- GitHub repository
- fetched 2026-10-19
- Tool list (tools/list), measured by us
- 2026-10-19
- Security signals, computed by us
- 2026-10-19
- Last checked against all sources
- 2026-10-19
The description, packages and links come from the maintainer's own published metadata. Token cost, security signals, freshness and the reading above are MCPSkillHub's own measurements; see the methodology. Spot an error? Request a correction or claim the listing.