Humaux Memory

Persistent long-term memory for AI agents over one remote MCP connection.

От сообществаРаботаетБез входаГлобальныйБесплатноТолько чтение

Что умеет

  • Memory Search: Semantic search across all memory layers. Automatically includes results from both your private memories and the shared public knowledge pool (built from high-quality internalized memor
  • Memory Enumerate: Enumerate the COMPLETE set of memories matching a structured filter — every rejected decision for a project, every lesson, every fix. Unlike memory_search (semantic top-k, returns on
  • Memory Grep: EXACT-SUBSTRING search (case-insensitive, deterministic) over the RAW title+content of every memory visible to your team (private + shared public pool). Zero LLM, zero embeddings: the pat

Какие данные видит

Нужен ли аккаунт

Не нужен: сервер работает без входа

Persistent long-term memory for AI agents over one remote MCP connection. Store decisions, lessons and documents; recall them semantically in any later session; enumerate complete sets with truthful totals instead of top-k guesses.

  • Memory — semantic store/recall with time-aware deduplication
  • Complete enumeration — "every decision", "all rejected approaches", with real totals and coverage
  • Per-project ledgers — current state, rejected paths, lessons, known issues
  • Shared task canvas — task state outside the context window, so another agent can resume mid-task
  • Knowledge graph — entities, relations and supersession links
  • Documents — upload and search your own files
  • Multi-agent coordination — task claims and resource locks with leases

One-click OAuth sign-in, no API key to paste. Bring your own LLM provider key for distillation. Free tier available.

Как подключить

Как подключить Humaux Memory к Claude

Адрес сервера, который нужно вставить в настройках Claude:

https://humaux-memory--luolimoa.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 Humaux Memory к ChatGPT

Адрес сервера, который нужно вставить в настройках ChatGPT:

https://humaux-memory--luolimoa.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 Humaux Memory к Cursor

Адрес сервера, который нужно вставить в настройках Cursor:

https://humaux-memory--luolimoa.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 (17)

Технические названия из tools/list. Нужны только разработчикам.

memory_searchSemantic search across all memory layers. Automatically includes results from both your private memories and the shared public knowledge pool (built from high-quality internalized memories via K2 pipeline). Supports time range (since/until), layer filter (L1-L4), and source pool filter.
memory_enumerateEnumerate the COMPLETE set of memories matching a structured filter — every rejected decision for a project, every lesson, every fix. Unlike memory_search (semantic top-k, returns only the most relevant few and never guarantees completeness), this walks Postgres directly and returns the FULL matching set, always reporting 'total' (true full-set size), 'coverage' and whether the answer was truncated. Omit 'limit' to get everything. Use this whenever the question is 'all / every / which ones have we…' rather than 'find the most relevant'.
memory_grepEXACT-SUBSTRING search (case-insensitive, deterministic) over the RAW title+content of every memory visible to your team (private + shared public pool). Zero LLM, zero embeddings: the pattern is matched literally, never as a regex or wildcard. Use this INSTEAD OF re-phrasing memory_search when semantic search came back empty or you are hunting an exact string — an identifier, an error message, a config key, a file/person name. Returns each matching memory with the matching lines (±1 line of context), a per-memory match count, and the TRUE total of matching memories. Superseded memories are included but flagged 已被更正·非当前. Follow up with memory_get(memory_id=…) for a full untruncated body.
memory_getFetch ONE memory by id — the COMPLETE original text (up to 50000 chars, versus the short snippets memory_search/memory_grep return) plus provenance metadata (layer, source_type, private/public origin, topic, category, created_at, lifecycle). Use it when a hit from memory_search / memory_enumerate / memory_grep was truncated or you need the exact full wording of a known id; it takes no query and does NO searching — for discovery use memory_search (semantic) or memory_grep (literal). Only memories visible to your team can be fetched. A superseded memory IS still returned for provenance, clearly flagged 已被更正·非当前 with the id that replaced it — do not treat flagged content as current fact.
memory_storeStore a memory, or correct an existing one. Omit 'id' to create a new memory (async). Pass 'id' to update that memory in place (content is re-embedded so search stays consistent). Pass 'supersedes' with the id of an older memory this one REPLACES: the old entry is marked superseded (kept for history, but demoted to near-zero in search and flagged '已被更正·非当前'), so a stale status can't outrank the correction.
memory_task_canvasShared cross-session task canvas. action='get' reads the whole team canvas; action='update' writes your plan/progress (mermaid + completed/next lists).
memory_deleteDelete a memory entry by its identifier.
code_searchSearch indexed code entities by function/class name.
code_indexIndex code you are developing into a project knowledge graph so you can later query its structure (functions, classes, call/import graph) instead of re-reading whole files — saves tokens. Pass the code via 'files' (recommended): the files you have open or just wrote. Returns immediately; use code_query(action='index_status') to check progress.
code_memory_linkLink a code entity to a memory entry.
doc_searchSearch uploaded documents.
memory_metaMeta queries over the memory system. action='persona' (L3 user profile), 'status' (system stats), 'skills' (RETIRED 2026-07-19 — the L4 skill layer was removed; recall procedures via memory_search instead. Returns empty.), 'related' (graph neighbors of a memory; pass query=id or title fragment), 'contradictions' (conflicting memories in the graph), 'list' (browse stored memories; pass limit), 'feedback' (report a skill outcome; pass skill_id + outcome=success|failure|partial, optional memory_ids/context).
code_queryQuery code relationships, impact analysis, or index status. Use action='relations' for calls/inherits, action='impact' for dependents, action='index_status' for indexing progress, action='history' for temporal evolution of an entity.
code_repo_mapGet a ranked, token-budgeted MAP of an indexed codebase — the most important functions/classes (by call-graph PageRank) as signatures + line numbers, NOT full source. Read this first to orient in a repo instead of reading whole files; then fetch specific symbols with code_query. Pass 'focus' (files/symbols you're working on) to bias the map toward them.
public_browseBrowse the shared public knowledge pool built from internalized high-quality memories. No args: returns topic list. Use topic='xxx' to filter by K2 topic. Use query='xxx' for semantic search within public knowledge only.
coord_taskMulti-agent task coordination for the team: submit work, claim it collision-free (lease-based), report progress, hand it off with context. Claim before working so parallel agents never duplicate effort; heartbeat before the lease expires or the task is requeued for others.
coord_lockAdvisory resource locks for parallel agents: claim a file/entity/resource BEFORE editing it so other agents avoid conflicts. Leases expire automatically if the holder dies; re-acquiring your own lock extends it.
Humaux Memory: подключить к Claude, ChatGPT, Cursor · Connectors.fun