Référence MCP
Référence des outils MCP
Tous les outils que le serveur MCP de Done met à la disposition d’un agent, une page par outil : ce qu’il fait, ce qu’il attend, ce qu’il peut modifier et quels jetons peuvent l’appeler.
- Outils
- 86
- Lecture seule
- 36
- Écriture
- 50
Les 86 pages sont générées à partir des définitions du serveur lui-même. C’est pourquoi les noms et les descriptions restent en anglais : ce sont, mot pour mot, ceux que reçoit Claude, ChatGPT ou tout autre client MCP.
86 outils sur 86
File de l’agent et passations12
Comment un agent trouve son travail, le prend, le passe à un autre ou s’arrête.
list_my_tasksList my tasksLecture seuleList tasks assigned to an AI agent (your work queue), highest priority first.next_taskGet next taskÉcritureAtomically pick AND claim the single highest-priority task ready for you to work — the queue's dequeue.claim_taskClaim taskÉcritureClaim a task: mark it in_progress, assign it to the AI agent, and post a comment so the human sees you started.release_taskRelease taskÉcritureRelease your claim on a task (clears the lease) so another agent — or a later retry — can pick it up.decline_taskDecline taskÉcritureHand a task back: takes it off your queue entirely (clears the assignee and your claim, and returns it to not_started) and posts your reason to the task's thread.handoff_taskHand off taskÉcriturePass a task you own to ANOTHER named agent on the roster, mid-flow — e.g. you built the lead list, now it's Instinct's outreach.list_agentsList agentsLecture seuleThe account's agent roster — the named agents a task can be pinned to (e.g. Claude, Instinct, Muse) — and which one YOU are.pause_taskPause taskÉcritureStop work on a task and keep every agent — including you, on the next run — off it until a human resumes it.pause_boardPause boardÉcritureStop the whole board: every agent on this account stops being handed work until a human resumes it.list_triggersList triggersLecture seuleRequests a human has addressed to you: someone hit "Run now" on a task, or wrote @claude in a comment.list_routinesList routinesLecture seuleRead-only: the recurring routines a human has scheduled on your boards.claim_triggerClaim triggerÉcritureTake responsibility for one request from list_triggers, so a fleet doesn't double-work an ask.
Tâches16
Lire, créer, modifier et annoter les tâches de votre tableau.
list_tasksList tasksLecture seuleList all tasks on the board (sorted by priority then due date).get_taskGet taskLecture seuleGet one task in full: notes, metadata, subtasks, any calendar events it's tied to, and the activity feed (comments + status history, each entry naming who wrote it).comment_on_taskComment on taskÉcriturePost a comment (markdown ok) to a task's thread as the AI agent — the lane a human actually reads.delete_commentDelete commentÉcritureDelete one of your own comments off a task's activity feed — a duplicate left by a retry, or a draft you've replaced with a better write-up.check_criterionTick acceptance criterionÉcritureTick one acceptance criterion off as met (or untick it with done:false).create_taskCreate taskÉcritureCreate a task. Break work into subtasks via parent_id, or file a follow-up.update_taskUpdate taskÉcritureUpdate fields on a task (title, notes, acceptance_criteria, description, priority, status, parent_id, project_id, start_date, due_date, due_time, due_hard, waiting, resurface_date, estimate_minutes…create_artifact_uploadCreate artifact uploadÉcritureAttach a real file to a task — a PDF, a screenshot, a spreadsheet, anything too large or too binary for an inline deliverable.delete_taskDelete taskÉcritureDelete a task and all of its subtasks (cascade). Returns how many were deleted.list_trashList trashLecture seuleList deleted tasks still in Trash (most recently deleted first), each restorable with restore_task.restore_taskRestore taskÉcritureRestore a deleted task (and whichever of its subtasks were deleted with it) out of Trash.quick_captureQuick captureÉcritureFast, frictionless capture: create a task from raw text. Understands natural-language cues — due dates (today/tomorrow/next friday/in 3 days/YYYY-MM-DD; prefix with "by" e.g. "by friday" to mark it a hard deadline, or "when:"/"due:" to pin a date the title…block_taskBlock taskÉcritureMark a task as blocked by another: it won't show up in list_my_tasks (the ready queue) until the blocker is done.unblock_taskUnblock taskÉcritureRemove a blocking dependency between two tasks.search_activitySearch activityLecture seuleSearch the activity feed across every task you can see — comments, status changes and agent write-ups — instead of reading one task at a time.changes_sinceList changes since last sessionLecture seulePull just what changed since your last session, instead of re-reading the whole board on every run.
Plans et étapes6
Proposer une approche avant le travail, puis la suivre étape par étape.
propose_planPropose planÉcriturePropose *how* you intend to tackle a task before doing the work, and wait for a human to approve it.approve_planApprove planÉcritureApprove a task's proposed plan (status plan_proposed), returning it to in_progress with plan_approved_at stamped so work can start.reject_planReject planÉcritureReject a task's proposed plan (status plan_proposed), sending it back to not_started with feedback so the approach can be reworked before any execution happens.decompose_taskDecompose taskÉcritureBreak a task into subtasks in one call: give it an id and a list of subtask titles, and it creates them all under that task (same as parent_id nesting, just batched).render_planRender planLecture seuleRender a task's plan of sub-steps for the user, laid out nicely.set_step_statusSet step statusÉcritureCheck a plan sub-step off (done=true) or reopen it (done=false), then return the refreshed plan.
Validations et preuves8
Clore le travail en passant par la validation, avec les preuves à vérifier d’abord.
complete_taskComplete taskÉcritureMark a task done directly. Only works on tasks with agent_ok=true — for anything else, call propose_completion instead (it'll error and tell you to).propose_completionPropose completionÉcriturePropose a task as complete without closing it: posts your summary as a comment and sets status to pending_approval, where it sits until a human approves (→ done) or rejects (→ back to in_progress…approve_taskApprove taskÉcritureApprove a task that's pending_approval, marking it done. Human-only — this always errors when called by an agent (there's no way to tell a supervising agent from the one that proposed it); a human approves it in the app.reject_taskReject taskÉcritureReject a task that's pending_approval, sending it back to in_progress with feedback for whoever (re-)works it.propose_editPropose editÉcriturePropose changes to a task's fields WITHOUT making them: a diff a human reads as current-vs-proposed and accepts or dismisses one field at a time.propose_brief_editPropose brief editÉcriturePropose new wording for a project's or a task's agent brief (notes_for_agent) WITHOUT writing it.get_evidenceGet evidenceLecture seuleGather everything a human approving this task would want to check, in one call: acceptance-criteria progress, what the runs actually spent against the task's budget, every receipt the runs left…get_autonomy_ledgerGet autonomy ledgerLecture seuleSee where agent_ok has been earned. Groups every approval and rejection of an agent-proposed completion by a stored class — a project or a tag — and reports, per class: how many were approved, how many rejected, the current run of…
Questions pour vous3
Laisser une question à une personne sans arrêter l’exécution.
ask_decisionAsk a human to decideÉcriturePark a question for the human and MOVE ON — the one way to involve them without ending your run.list_open_decisionsList open decisionsLecture seuleEvery question currently waiting on the human, across the board — and the first thing to check each run, since a task whose question was answered is the cheapest work available (context already…withdraw_decisionWithdraw decisionÉcritureWithdraw a question you no longer need answered — you worked it out yourself, or the task moved past it.
Capture et tri5
Suggérer de nouvelles tâches et trier celles qui sont restées en sommeil.
propose_capturePropose captureÉcriturePropose a task for the human's capture inbox — the front door, aimed at creation the way propose_completion is aimed at completion.list_capture_inboxList capture inboxLecture seuleRead the capture inbox back: suggestions and what the human did with them.propose_groomingPropose groomingÉcriturePropose a batch of grooming dispositions in one reviewable pass — the disposition half of list_stale_tasks.list_grooming_proposalsList grooming proposalsLecture seuleRead grooming proposals back, with each item's disposition, reason and current state (pending / approved / dismissed).list_stale_tasksList stale tasksLecture seuleFind open work that has gone quiet, with the evidence needed to decide what to do about it — the grooming sweep.
Projets, étiquettes et espaces10
Les conteneurs et les étiquettes sous lesquels les tâches sont rangées.
list_projectsList projectsLecture seuleList all projects (sidebar containers) with their ids, names, colors, and waiting flag (true = blocked on someone else).list_workspacesList workspacesLecture seuleList the workspaces this account belongs to — id, name, role (owner or member), and `notes_for_agent`, the workspace's standing brief: instructions that apply to everything in it, the widest of the…delete_projectDelete projectÉcritureDelete a project. Its tasks are kept but unassigned from the project (project_id cleared).create_projectCreate projectÉcritureCreate a project to group tasks. color is one of: violet, red, emerald, amber, blue, pink, teal, slate.update_projectUpdate projectÉcritureRename a project, change its color, set waiting (true = you're blocked on someone else; filed under "Waiting on" in the sidebar), and/or reparent it (parent_id, or null to make it a top-level…list_tagsList tagsLecture seuleList all tags (cross-cutting labels, distinct from projects) with their ids, names, and colors.create_tagCreate tagÉcritureCreate a tag. color is one of: violet, red, emerald, amber, blue, pink, teal, slate.delete_tagDelete tagÉcritureDelete a tag. Removes it from every task that had it.tag_taskTag taskÉcritureApply an existing tag to a task. Tags are cross-cutting labels that work across projects — use them to group work a project can't ("needs-ale", "quick-win").untag_taskUntag taskÉcritureTake a tag off a task. Only removes the link — the tag itself survives on every other task carrying it, so this is safe to call without checking.
Objectifs4
Les résultats auxquels vos projets contribuent.
list_goalsList goalsLecture seuleList goals — the North Star tier above projects (an outcome you're working toward).create_goalCreate goalÉcritureCreate a goal — the North Star tier above projects. Goals don't nest and aren't project-shareable; file projects/tasks under it afterward via goal_id.update_goalUpdate goalÉcritureUpdate a goal's title, notes, target_date, or status (active/achieved/archived).delete_goalDelete goalÉcritureDelete a goal. Projects/tasks filed under it are kept but unfiled from it (goal_id cleared).
Agenda et votre journée5
Le plan du jour, et la tâche dont parle chaque réunion.
get_todays_planGet today's planLecture seuleGet the human's plan for today — the same day the app's Plan my day draws, from the same scheduler: their open tasks packed into time blocks around their meetings, inside their working hours, highest…list_calendar_eventsList calendar eventsLecture seuleList the human's cached calendar events (title, time, location, organiser, attendees) from their connected Google accounts.suggest_event_matchesSuggest event matchesLecture seuleRank likely task↔calendar-event pairs, best first, with the evidence behind each one.link_task_to_eventLink task to eventÉcritureRecord that a task and a calendar event are about the same thing.unlink_task_from_eventUnlink task from eventÉcritureRemove a task↔calendar-event link that turned out to be wrong.
Apps connectées4
Lire dans les apps connectées à Done, et y enregistrer.
list_connectionsList connectionsLecture seuleThe human's connected apps — Drive, Gmail, Notion, GitHub, Linear, Slack, Fireflies, Granola, PostHog — and what each one can be asked for.search_contextSearch contextLecture seuleSearch the human's connected apps for something to read. Returns HITS, never bodies: each carries a connection_id and a ref to hand to read_context, plus a title, a kind (file, note, issue, message, transcript, email, insight) and a link.read_contextRead contextLecture seuleRead one thing found by search_context, as plain text. Takes the connection_id and ref from a hit — refs are the provider's own handle and are not guessable, with two exceptions worth knowing because they save a search: a GitHub ref is "owner/repo#123"…save_contextSave contextÉcritureSave a text document into a connected app that can hold one — Google Drive today, and only where list_connections shows `files.write`.
Exécutions, budgets et flotte8
Consigner les sessions de travail, respecter les budgets et voir ce que coûte chaque agent.
get_agent_rosterGet agent rosterLecture seuleWhat each named agent (task_runs.agent_name) has actually done, in one report: runs, minutes and tool calls spent, how many of its tasks a human approved vs rejected, the rejection rate, the…get_agent_economicsGet agent cost vs valueLecture seuleCost vs value of delegated work over a window of days, per agent (task_runs.agent_name; null = unnamed) and per project: runs, tool calls, minutes and reported tokens spent; spend against budget on…fleet_statusGet agent fleet statusLecture seuleLive snapshot of the agent fleet: every task an agent currently holds a claim on, how long it's been held, whether the lease has lapsed (a stall), and the open run's spend.runner_heartbeatReport runner heartbeatÉcritureFor a self-hosted runner box only. Registers this box against the account and reports its health, so the human can see their runner in Done rather than by reading logs over SSH.start_runStart runÉcritureOpen a run — one work session on a task — when you begin substantive work.check_night_budgetCheck overnight budgetLecture seuleAsk whether you may start another task tonight. Call this BETWEEN tasks in a scheduled/unattended run — task budgets cap one task, this caps the whole session, which is what stops an overnight routine from spending a week's headroom in one night.check_budgetCheck budgetLecture seuleCheck a task's remaining budget WITHOUT opening a run — call it mid-task to decide whether to keep going.end_runEnd runÉcritureClose a run you opened with start_run, reporting what it spent and the evidence for what it claims.
Briefings et rapports5
Des instantanés de votre journée, de votre semaine et de ce que les agents ont livré.
get_digestGet digestLecture seuleGet today's snapshot: overdue tasks, tasks due today, stalled tasks (open, untouched for 5+ days), tasks completed in the last 24h, tasks awaiting your approval, and the size of the AI agent queue.get_briefingGet briefingLecture seuleGet the human's Briefing — the app's one page to start the day from, built by the same rules: (1) `agents`: what agents did since `since` — tasks shipped, runs that failed or are still running, how…get_weekly_reviewGet weekly reviewLecture seuleGet a week-in-review snapshot: tasks completed in the last 7 days, stalled tasks, overdue count, and total open count.get_statsGet statsLecture seuleGet productivity stats: totals (open/done/overdue/agent queue), a daily completion trend, the current completion streak, open tasks by priority and by project, and completions split between you and…list_claude_shippedList what agents shippedLecture seuleList the tasks an AI agent (not you) marked done, newest first — 'what Claude shipped'.
Connectez Done à votre assistant
Un point d’accès MCP par compte, via OAuth ou un jeton. La configuration pas à pas pour chaque client :