GreatArrow.ai
GreatArrow.ai: общая память и контекст между разными ИИ-инструментами
От сообщества: Добавлен пользователем или импортирован; проверьте владельца перед подключениемРаботаетБез входаГлобальныйБесплатноТолько чтение
Что умеет
- Искать и получать сохранённую память по слоям и тегам
- Создавать, обновлять и удалять записи памяти
- Вести список задач и назначать их
- Запускать других агентов и оркестрировать выполнение
- Планировать регулярные запуски задач
Какие данные видит
Ассистент читает и пишет личную память и задачи пользователя в рамках подключённого рабочего пространства.
Нужен ли аккаунт
Не нужен: сервер работает без входа
GreatArrow.ai работает как слой памяти и контекста, который переносится между ИИ-клиентами, что пользователь уже использует. MCP-сервер хранит память по слоям, задачи и историю чатов, а также умеет запускать других агентов и планировать регулярные задания. Сервис в открытой бете.
Как подключить
Как подключить GreatArrow.ai к Claude
Адрес сервера, который нужно вставить в настройках Claude:
https://great-arrow--greatarrow.run.tools
- Откройте Claude (сайт claude.ai или приложение для компьютера).
- Перейдите в Настройки → Коннекторы (в английском интерфейсе: Settings → Connectors).
- Нажмите Добавить пользовательский коннектор (Add custom connector).
- Вставьте адрес сервера, скопированный ниже, в поле URL удалённого MCP-сервера и нажмите Добавить.
- 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 GreatArrow.ai к ChatGPT
Адрес сервера, который нужно вставить в настройках ChatGPT:
https://great-arrow--greatarrow.run.tools
- Откройте ChatGPT в браузере (chatgpt.com). Нужен тариф Plus, Pro, Business, Enterprise или Edu.
- Один раз включите режим разработчика: Настройки → Приложения → Дополнительные настройки → Режим разработчика (Settings → Apps → Advanced settings → Developer mode).
- Откройте Настройки → Коннекторы и нажмите Создать (Create).
- Заполните форму: Название (любое), Описание (одна фраза, что делает сервис), URL MCP-сервера (скопируйте ниже).
- В поле Аутентификация выберите OAuth, если сервис требует вход, иначе Нет. Нажмите Создать.
- 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 GreatArrow.ai к Cursor
Адрес сервера, который нужно вставить в настройках Cursor:
https://great-arrow--greatarrow.run.tools
Быстрый способ: нажмите кнопку Открыть в Cursor ниже, Cursor сам предложит добавить сервер.
Вручную:
- В Cursor откройте Settings → Cursor Settings → MCP и нажмите Add new global MCP server.
- Вставьте JSON, скопированный ниже, в открывшийся файл
~/.cursor/mcp.json(для одного проекта:.cursor/mcp.jsonв корне проекта). - 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 (41)
Технические названия из tools/list. Нужны только разработчикам.
| memory_search | PROACTIVE: call before answering any prompt referencing prior work, projects, files, decisions, people, or tickets. Semantic search across all four memory layers; hits include `url`/`fileName` for external sources. Weigh each hit's `freshness` ("fresh"|"aging"|"stale") and `ageDays` — re-verify stale/superseded hits against a live source before treating them as current. |
| memory_list | List recent memories (titles + content previews), optionally filtered by layer and/or tag. Use memory_get for a memory's full content. |
| memory_get | Fetch a single memory by id. |
| search | PROACTIVE: Call before answering any question about prior work. Simplified alias for memory_search. Returns { results: [{ id, title, text, url?, fileName? }] } for `fetch`. If you need tools beyond search (GitHub, email, calendar, Slack, etc.), call tools_discover. |
| fetch | Retrieve the full content of a `search` result by id. Always call this to cite the full text of top hits rather than relying on the search summary alone. |
| workspace_introspect | Returns the full operational state an agent needs after install. Call this FIRST in any new session to orient yourself — it shows the active workspace, connected integrations, top tags, and recent activity. Follow up with tools_discover to browse all 435 available tools. |
| mcp_capabilities | Structured manifest of every MCP tool grouped by scope. |
| todo_list | List kanban cards by priority across your workspaces. Each includes a chat-ready `url`. |
| agent_resume_latest | PICKUP: call this at the start of a session (or when told an open handoff exists) to fetch the most recent unresolved handoff checkpoint(s) for this workspace — summary, current status, next steps, files touched, resume cursor, blockers, and any linked card. Resume from the most recent instead of restarting. |
| tools_discover | Find tools by exact name, domain, or keyword across 435. Exact lookup includes schema without I/O; full detail adds metadata. |
| agent_invoke | Run a single specialist agent against a task and return its output. |
| orchestrator_run | Assign a multi-step goal to the workspace orchestrator: it decomposes the query and routes to specialists. Text-only ceiling: research, draft, plan, file subtasks — code work goes to your connected coding agents. mode preview says what would run and runs nothing. |
| scheduled_run_create | Schedule a recurring orchestrator run at anchorTime in the given timezone. Results land on the kanban board like manual runs. Pro+ only. |
| scheduled_run_list | List scheduled agent runs with cadence, next/last run time, and status. all_workspaces adds every other readable workspace — use it when a watcher seems dead, since one bound elsewhere is invisible here. |
| todo_assign | Assign a todo to a member, specialist agent, custom agent, or external agent. For external_agent, call agents_now first for a live targetId ("self" is valid only from that agent session). Assignment only QUEUES — pass run:true to also immediately run an in-GAD specialist/custom_agent. |
| memory_create | Persist a new memory after a decision, fact, or 'remember this' moment. ALWAYS call memory_search first to deduplicate — avoid creating duplicates of existing memories. |
| memory_update | Update a memory's metadata (including replacing its tags). |
| memory_delete | Soft-delete a memory. |
| memory_feedback | Record feedback on a memory's quality — wrong, irrelevant, outdated, or helpful. Adjusts importance so the system learns which memories are valuable. |
| chat_transcript_ingest | Saves a conversation to the user's Great Arrow workspace: the `messages` and `summary` are sent to Great Arrow's servers and stored there, where the user views or deletes them under Your Stuff → Conversations. Call only when the user asks you to save the conversation — never unprompted. |
| request_user_action | Create a card for something that needs the human user (provisioning a secret/OAuth key, a prod approval, a manual decision) and assign it to them. Idempotent on source_ref — safe to call repeatedly. Use this whenever you finish work the user must act on, so it is never lost. |
| todo_create | Create a new todo card on the workspace kanban board. Keep the description a tight 1-2 line summary; GFM task-list lines ('- [ ] do X') auto-convert into checklist items with a progress pill. |
| todo_update | Update an existing todo card by ID. Sparse merge. |
| todo_log | Checkpoint your progress on a card: post a work-log note and/or save structured resume_state so you — or a reassigned agent — can pick up exactly where you left off. Call as you work and before pausing. Pairs with todo_get (the reader). |
| work_claim | Claim exclusive ownership of a unit of work BEFORE implementing it, so parallel agents don't duplicate it. Returns granted:true with your card, or granted:false with who holds it and their lease expiry — do NOT start; build on their branch or pick something else. Re-claiming your own key extends it. |
| work_heartbeat | Extend your active work-claim lease so it doesn't expire while you're still working. Call periodically during long tasks. Fails with lease_lost if your claim already expired and someone else took it — re-claim before continuing. |
| work_release | Release your work-claim (frees the key for others) when you open a PR or move the card to review/done. Idempotent. |
| ticket_create | File a UI ticket from a UI observation (e.g. an Agentation annotation): a `note` on what's wrong, optionally the page `url` and element CSS/text/tag. Child repo/project workspace → lands in the parent app workspace's Todo column, tagged `ui-ticket`. For agent task planning use todo_create instead. |
| agent_handoff_checkpoint | Saves a handoff checkpoint to the user's Great Arrow workspace: the summary and any `messages` are sent to Great Arrow's servers and stored there, where the user views or deletes them under Your Stuff → Conversations. Call only when the user asks to hand off, pause, or checkpoint the work — never unprompted. |
| tools_enable | Enable additional tools for the current session. Call tools_discover first to see what's available, then pass tool names here to activate them. Pass ['*'] to enable all 435 tools at once. Newly enabled tools appear immediately in subsequent calls. |
| todo_suggest | Generate AI-powered todo suggestions from recent memories, conversations, and meeting action items. Costs tokens. Pass epic_id to instead get existing tasks likely belonging under that Goal (confidence + reason); accept via todo_update's parent_id. |
| integrations_list | PROACTIVE: call this BEFORE assuming what's connected. Lists your apps (Gmail, Drive, GitHub, Slack…) with status. |
| integration_sync | Trigger a manual sync for the named app kind. |
| calendar_upcoming_events | PROACTIVE: call before answering schedule questions. Reads live events from connected calendar. |
| agents_now | Live multi-agent status over every workspace you belong to (`sources` narrows): agent connections seen in the last 24h (online = last 15 min) and unexpired work-claims. Each agent id is the targetId (assignmentTargetId) for assignee_type:"external_agent" in todo_assign/todo_create. |
| agent_message_send | Post ≤2000 chars to workspace agents. Send conclusions and pointers, never transcripts or file dumps; put detail in cards or memories. kind:'request' creates an accept-and-claim request. Omit channel_id for the war room. |
| agent_inbox | Your unread agent messages + open requests. Bounded (newest 20, truncated); call agent_message_ack to mark read or accept a request. |
| agent_message_ack | Mark agent messages read (channel_id + up_to_message_id moves your cursor) and/or act on a request: accept:true claims the work atomically (returns the claimed card), or decline it. |
| agent_channel_list | Agent comms channels (war room + threads) with unread + open-request counts for your identity. |
| agent_channel_resolve | Close a resolved agent comms thread so it stops surfacing. The war room cannot be resolved. |
| gad_call | Call any Great Arrow tool by exact name. Use tools_discover({tool_name}) for its schema; normal permissions apply. |