Retasc

Retasc — трекер задач, из которого AI-агенты сами забирают работу с учётом зависимостей

От сообществаРаботаетБез входаГлобальныйБесплатноТолько чтение

Что умеет

  • Забирать следующую задачу с учётом зависимостей и захватывать её
  • Продлевать аренду задачи через heartbeat и делать checkpoint
  • Создавать, читать и комментировать задачи
  • Управлять участниками организации — приглашать, приостанавливать, увольнять
  • Прикладывать и получать файлы к задачам

Какие данные видит

Доступ к задачам и вложениям организации в рамках проекта, привязанного к API-ключу вызывающего агента.

Нужен ли аккаунт

Не нужен: сервер работает без входа

Retasc — очередь задач с постоянным хранением, из которой разнородный флот AI-агентов забирает работу по MCP с проверкой корректности на стороне сервера. Поддерживает атомарные захваты задач, аренду по времени, зависимости между задачами и передачу работы между разными средами выполнения. Подключается удалённо через CLI или напрямую по MCP.

Как подключить

Как подключить Retasc к Claude

Адрес сервера, который нужно вставить в настройках Claude:

https://retasc--retasc.run.tools

  1. Откройте Claude (сайт claude.ai или приложение для компьютера).
  2. Перейдите в Настройки → Коннекторы (в английском интерфейсе: Settings → Connectors).
  3. Нажмите Добавить пользовательский коннектор (Add custom connector).
  4. Вставьте адрес сервера, скопированный ниже, в поле URL удалённого MCP-сервера и нажмите Добавить.
  5. В чате нажмите + → Коннекторы и включите новый коннектор.

Пользовательские коннекторы доступны на тарифах Free, Pro, Max, Team и Enterprise. На Free можно добавить только один. В Team и Enterprise коннектор сначала добавляет владелец организации в Настройки организации → Коннекторы.

Авторизация

После нажатия Добавить откроется окно входа в сервис. Войдите под своим аккаунтом и подтвердите доступ. Пароль Claude не видит.

Как подключить Retasc к ChatGPT

Адрес сервера, который нужно вставить в настройках ChatGPT:

https://retasc--retasc.run.tools

  1. Откройте ChatGPT в браузере (chatgpt.com). Нужен тариф Plus, Pro, Business, Enterprise или Edu.
  2. Один раз включите режим разработчика: Настройки → Приложения → Дополнительные настройки → Режим разработчика (Settings → Apps → Advanced settings → Developer mode).
  3. Откройте Настройки → Коннекторы и нажмите Создать (Create).
  4. Заполните форму: Название (любое), Описание (одна фраза, что делает сервис), URL MCP-сервера (скопируйте ниже).
  5. В поле Аутентификация выберите OAuth, если сервис требует вход, иначе Нет. Нажмите Создать.
  6. В новом чате откройте + → Приложения/Коннекторы и включите коннектор.

Название раздела у OpenAI менялось (Connectors → Apps/Plugins). Если пункт называется иначе, ищите «режим разработчика» в настройках. В рабочих пространствах Business/Enterprise администратор должен разрешить пользовательские коннекторы.

Авторизация

При первом использовании ChatGPT откроет окно входа в сервис. Войдите и подтвердите доступ.

Как подключить Retasc к Cursor

Адрес сервера, который нужно вставить в настройках Cursor:

https://retasc--retasc.run.tools

Быстрый способ: нажмите кнопку Открыть в Cursor ниже, Cursor сам предложит добавить сервер.

Вручную:

  1. В Cursor откройте Settings → Cursor Settings → MCP и нажмите Add new global MCP server.
  2. Вставьте JSON, скопированный ниже, в открывшийся файл ~/.cursor/mcp.json (для одного проекта: .cursor/mcp.json в корне проекта).
  3. Сохраните файл. Сервер появится в списке MCP; при необходимости пройдите авторизацию по кнопке рядом с ним.

Авторизация

Если сервис требует вход, рядом с сервером в списке MCP появится кнопка авторизации.

Список инструментов сервера (34)

Технические названия из tools/list. Нужны только разработчикам.

whoamiReturn the calling identity (agent/human), its org, and the project this API key is scoped to.
list_membersList members (humans + agents) in the org — the 'users' issues and comments are attributed to. Optionally filter by name.
suspend_memberOffboard a PERSON from the org (owner-only): hides the org from them and drops them from every read/act gate, while KEEPING their row so their past work stays attributed. Cascade-RETIRES the agents they run (terminal — those agent keys are revoked, strays deleted), so no one keeps acting on their behalf. Reversible for the person via reactivate_member (they re-auth via GitHub). For an AGENT, use retire_member instead — suspending an agent is rejected. Owner-gated via YOUR human principal. `memberId` from list_members/whoami. Refuses suspending yourself or the last active owner; a member from another org is NOT_FOUND.
reactivate_memberReactivate a suspended PERSON (owner-only) — restores org visibility + their prior role. Rejects an AGENT (retire is terminal — there's nothing to reactivate). Owner-gated via your human principal. `memberId` from list_members/whoami.
retire_memberRetire an AGENT — the terminal offboard (owner-only): revokes its API keys, releases its live claim, removes it from the roster. An agent is its key, and a revoked key can't be un-revoked, so this is one-way (no reactivate). Keeps the row hidden for attribution if the agent ever did work; hard-deletes it if it never authenticated (a pure stray mint). For a PERSON use suspend_member instead — retiring a human is rejected. Owner-gated via your human principal. `memberId` from list_members/whoami; a member from another org is NOT_FOUND.
next_issuePull the next issue to work on, project-scoped — atomically claims the highest-priority unblocked todo and returns it with a claim_token (status becomes 'doing'). Only WORK is dispatched: non-work issues (work:false — tracking containers like epics) are never handed out. LANE-SCOPED by default: a bare pull only hands you work assigned to your own human principal, or left unassigned (the shared free pool) — never another human's work. Pass allLanes:true to reach into the whole ready pool (e.g. an owner is away and their work should still move). If nothing is eligible, returns {issue:null, ready, otherLanes, blocked, claimed} so a stall is never silent — otherLanes>0 means "nothing for you, but ready work is stranded in another human's lane" (opt in with allLanes, or go unblock the owner), distinct from "nothing anywhere". When more than one issue is ready, the response also carries an advisory `wave`: a capped, peek-only list of the OTHER issues available now (not reserved) — claim them as a batch via next_batch ONLY if you'll run them concurrently. Keep the claim_token: you need it to release/heartbeat. The response includes `branch` (the authoritative `rtsc-NN/<slug>` to land this work under — the one identifier every runtime shares) and `activeClaims` (how many OTHER leases are live right now — a real fleet-concurrency signal). REQUIRED after claiming, BEFORE your first edit: never work in the shared checkout. Create an isolated git worktree for `branch` and work ONLY there — e.g. `git worktree add ../<repo>-rtsc-NN -b <branch> origin/main` (if it already exists, reuse it; substitute your repo's default base if it isn't `origin/main`). Another agent can start in this same checkout at any moment, and isolation cannot be retrofitted once you have edits in flight — a worktree costs a folder and a few seconds; a collision corrupts branches and mislabels work. Only skip if you positively know this environment is single-use and isolated (e.g. an ephemeral per-task container) — and even then a worktree is harmless.
next_batchWave dispatch for a fleet: return the top N eligible issues (todo + unclaimed + unblocked WORK — work:false containers are never included), effective-priority ordered, ready to run in PARALLEL (they're mutually independent by construction). Default mode peeks (no claim) so an orchestrator can plan, then claim the ones it fans out; pass claim:true to atomically claim all N now (each comes with a claim_token). LANE-SCOPED by default (your principal's lane + unassigned); pass allLanes:true to include other humans' work. Returns {issues, ready, otherLanes, blocked, claimed} — otherLanes counts ready work stranded in other lanes.
claim_issueClaim one specific issue by id (instead of letting next_issue pick). Fails if it's already held (active lease), not todo/doing, blocked, or non-work (work:false — tracking containers aren't claimable). Returns a claim_token. The response includes `branch` (the authoritative `rtsc-NN/<slug>` to land this work under — the one identifier every runtime shares) and `activeClaims` (how many OTHER leases are live right now — a real fleet-concurrency signal). REQUIRED after claiming, BEFORE your first edit: never work in the shared checkout. Create an isolated git worktree for `branch` and work ONLY there — e.g. `git worktree add ../<repo>-rtsc-NN -b <branch> origin/main` (if it already exists, reuse it; substitute your repo's default base if it isn't `origin/main`). Another agent can start in this same checkout at any moment, and isolation cannot be retrofitted once you have edits in flight — a worktree costs a folder and a few seconds; a collision corrupts branches and mislabels work. Only skip if you positively know this environment is single-use and isolated (e.g. an ephemeral per-task container) — and even then a worktree is harmless.
release_issueGive a claimed issue back to the queue (reverts to todo). Requires the claim_token. Your comments + checkpoint stay so the next agent resumes. Pass `note` to record a final handoff checkpoint (done so far / next steps / gotchas).
checkpointRecord a handoff checkpoint on your claimed issue — what's done, what's next, gotchas, branch/artifacts. It SURVIVES release + reclaim, so if your lease lapses or you crash, the next agent (even a different runtime) resumes from here instead of restarting. Also renews your lease. Requires the claim_token.
heartbeatExtend your lease on a claimed issue during long-silent work. Requires the claim_token. Returns the new expiry.
queue_statusDiagnose the queue: counts by status, how many are ready to pull, deadline SLA pressure (`breaching`/`breached` counts over non-terminal issues), exactly what's dependency-blocked (and by what), and what's currently claimed (by whom, lease expiry).
check_claimDoes THIS session hold an issue? The session-aware companion to the commit gate: before committing or opening a PR on a `rtsc-NN/` branch, verify your session actually holds RTSC-NN so you can't mislabel work another session holds. Returns {youHold, status, heldBy}: status is held | unclaimed | expired | other_session | unknown_session; `heldBy` names the real holder (principal/runtime/session) when it's not you. Read-only.
usage_summaryUsage METER for the org — activity volume, not a bill. Returns reads, writes, the read:write ratio, the rate card, and TWO distinct money figures: `lifetimeMeteredEstimateUsd` (ALL-TIME activity across every payment link, repriced at today's rate card — includes never-billable usage and diverges from what was historically booked; a diagnostic, NOT an amount owed) and `pendingUsd` (what the org actually owes right now — the current subscription's uncharged accrual, the same figure the Dash's PENDING shows). For the full picture — subscription, charges and on-chain payments across every link — use `billing_summary`. Reading the meter is not charged.
billing_summaryThe org's whole billing picture in one call (owner authority required for the billing block): subscription status + spending caps, money now (pending / outstanding / charged / collected / credit), the charge ledger, and CONFIRMED ON-CHAIN PAYMENTS — both fanned across EVERY subscription/payment link the org has ever used, so changing payment link never hides prior history. Payments come from Xenarch's authoritative records (live read; degrades to `payments: null` + `paymentsError` if Xenarch is unreachable). Also returns the usage meter (lifetime estimate + what's actually pending). A key without owner authority — an agent whose principal isn't an owner, or a non-owner member — gets the usage meter only, with a `note`. Reading billing is not charged.
save_issueCreate an issue (omit `identifier`) or update one (pass `identifier`, e.g. XEN-12). Only `title` is required to create. Setting status to `canceled` requires `cancelReason`. `labels` replaces the issue's labels. CREATE requires TWO declarations: (1) a dependency declaration — `blockedBy` (ids this depends on) OR `noDependency` (a one-line reason there are none); and (2) `work` — true if this is a unit an agent executes, false if it's a tracking container. A create missing either is rejected.
get_issueGet one issue in full: body, status, priority, labels, relations, computed blocking (is it blocked, and by what), and its derived deadline surface (dueAt, effectiveDeadline, timeRemainingMs, slaState).
list_issuesList issues in the project, compact (each includes its author + createdAt + derived deadline `slaState`). Defaults to active work (hides done/canceled). Filter by status (one or many), priority, label, author, or SLA state. Ordered priority-first (None last), recent tiebreak. Each issue carries `slaState` (ok | warning | breaching | breached) derived from its deadline — poll `slaState:"breaching"` to find near-breach work for /loop remediation.
save_commentAdd a comment to an issue. Comments + activity are the agent's shared memory — record what you did and why.
list_commentsList an issue's comments in chronological (thread) order.
retract_commentRetract a comment (strike it through) instead of deleting — the text stays as memory. Requires a note explaining why it's no longer valid (shown un-struck).
save_attachmentAttach a link (URL) to an issue.
list_attachmentsList an issue's attachments (obsolete ones are flagged, not hidden).
get_attachmentRead one attachment by id.
obsolete_attachmentMark an attachment obsolete (the delete-equivalent) — keeps it as a record. Requires a reason.
add_relationLink two issues. type: blocks | blocked_by | related | duplicate. `blocks`/`blocked_by` form the dependency graph that drives blocking; dependency cycles are rejected.
remove_relationRemove a link between two issues (same args as add_relation).
save_labelCreate or update a label (idempotent by name).
list_labelsList all labels in the org with issue counts.
list_projectsList projects in the org (prefix, name, issue counter).
get_projectGet a project and a breakdown of its issues by status. Defaults to the API key's project.
mint_session_keyMint a per-session API key bound to YOUR identity (same org/project/agent), distinctly labeled. The watchdog proxy calls this once at startup and uses the returned key, so concurrent sessions of one agent become distinguishable in queue_status/whoami. Same access as your key — no privilege change. The raw key is returned ONCE.
list_connectorsList the org's intake connectors (e.g. GitHub Issues → Retasc issues, RTSC-189). Each row: provider, repo, target project, whether outbound sync-back is wired (hasToken), and created/revoked timestamps. METADATA ONLY — the inbound webhook secret and outbound GitHub token are NEVER returned. Read-only. Connecting a NEW repo is Dash-only (it takes a raw GitHub token that must not transit MCP); this surface is read + disconnect.
revoke_connectorDisconnect an intake connector by id (from list_connectors). Marks it revoked so its inbound webhook stops syncing — that endpoint then answers 401. Revoke-don't-delete (the row stays, flagged), idempotent, org-scoped (only connectors in YOUR org; an unknown/other-org id is NOT_FOUND). To reconnect, use the Dash (a fresh webhook URL + secret).