База знаний с автопилотом для вашего агента кодинга.
Установите один раз. С этого момента ваш агент сам выстраивает структурированную базу знаний вашего проекта — пока вы пишете код. Никаких промптов, которые нужно помнить, никаких вызовов remember(), никакого API для изучения.
Один бинарный файл Go. Встроенная SQLite. Никакого облака. Никакого Docker. Никакого Python. Никакой среды выполнения Node.
Agent (Claude Code / Cursor / Kiro / Gemini CLI / ...)
↓ MCP stdio
mnemos serve
↓
Auto-compiled knowledge base (~/.mnemos/mnemos.db)Любой сервер памяти хранит текст. Mnemos компилирует базу знаний.
В то время как другие серверы ожидают, что вы (или тщательно настроенный промпт) решите, когда сохранять и когда извлекать, mnemos запускает полноценный конвейер в фоновом режиме:
Agent action → mnemos auto-pipeline:
├── Quality gate (reject/rewrite low-value content)
├── 3-tier dedup (hash → fuzzy → semantic)
├── Auto-summarize (extractive, fast; LLM if available)
├── File linking (extract identifiers, link to code)
├── Type classification (episodic / long_term / semantic / working)
├── Quality scoring (for retrieval ranking)
└── Decay scheduling (so knowledge base stays relevant)
Retrieval:
├── Hybrid search (FTS5 + optional semantic + RRF)
├── File-overlap boost (memories about active files rank higher)
├── MMR diversity (kill redundant results)
├── Adaptive packing (full content or summary based on budget)
└── Token-budget cap (always fits in context)Вы не вызываете ничего из этого. Ваш агент не вызывает ничего из этого. Хуки запускают это автоматически при начале сессии, отправке промпта и завершении сессии.
Related MCP server: Gingugu
Уровень 1 — транспорт MCP. Стандартный MCP-сервер, stdio, работает с любым MCP-клиентом.
Уровень 2 — хуки автопилота. Одна команда (mnemos setup claude) настраивает хуки + управление + конфигурацию MCP. Начало сессии автоматически внедряет релевантный контекст. Отправка промпта автоматически выполняет поиск при смене темы. Завершение сессии проверяет полноту данных.
Уровень 3 — автоматически компилируемая база знаний. Контроль качества, трехуровневая дедупликация, авто-суммаризация, связывание файлов, сборка контекста MMR — все автоматически. Вы никогда не запускаете их вручную. Включает пассивный фоновый демон, который постоянно обнаруживает устаревание, противоречия и недостающие связи в вашей базе памяти.
Mem0 | Zep/Graphiti | engram | OMEGA | mnemos | |
MCP-native | ✓ | ✓ | ✓ | ✓ | ✓ |
Один бинарный файл, без зависимостей | — | — | ✓ | — | ✓ |
Без облака / локально-ориентированный | частично | — | ✓ | ✓ | ✓ |
Настройка автопилота одной командой | — | — | — | — | ✓ |
Автоматический контроль качества | — | — | — | — | ✓ |
Авто-суммаризация | — | — | — | — | ✓ |
Авто-связывание файлов (git-aware) | — | — | — | — | ✓ |
Сборка контекста MMR | — | — | — | — | ✓ |
Пассивный фоновый демон | — | — | — | — | ✓ |
Временной граф знаний | — | ✓ | — | — | частично (затухание + замещение) |
Стоимость хостинга | $0-облако | ~$50/мес (Neo4j) | $0 | $0 | $0 |
Mnemos не пытается быть Zep — это другой подход. Zep — лучшее решение, если вам нужно временное рассуждение о бизнес-фактах и у вас есть корпоративная инфраструктура. Mnemos — лучшее решение, если вы пользователь агента кодинга и хотите базу знаний с автопилотом, которая работает сама на вашем ноутбуке.
# Homebrew (macOS / Linux)
brew install s60yucca/tap/mnemos && mnemos setup claude
# curl (verify mnemos.dev is live before using)
curl -fsSL https://mnemos.dev/install.sh | bash && mnemos setup claude
# npm (coming in v1.2)
# npx mnemos setup claude
# Build from source (requires Go 1.23+)
git clone https://github.com/s60yucca/mnemos
cd mnemos && make buildЗамените claude на cursor, kiro или gemini-cli. Перезапустите клиент. С этого момента работает автопилот.
mnemos setup <client> записывает:
Файл управления (CLAUDE.md, .cursorrules, .kiro/steering/mnemos.md) — говорит агенту, что стоит сохранять
Конфигурацию хуков (.claude/hooks.json или эквивалент) — настраивает события жизненного цикла
Конфигурацию MCP (.mcp.json) — регистрирует mnemos serve как поставщика инструментов
Три хука запускаются автоматически:
Начало сессии → mnemos hook session-start
Собирает релевантные воспоминания в рамках лимита токенов (диверсифицированные по MMR, с приоритетом файлов). Внедряет в контекст. Холодный старт < 200 мс.
Отправка промпта → mnemos hook prompt-submit
Обнаруживает изменения темы + намерения. Автоматически ищет в базе знаний, когда сдвиг значим. Соблюдает кулдаун, чтобы избежать шума.
Завершение сессии → mnemos hook session-end
Проверяет, была ли зафиксирована долгосрочная память. Опционально сохраняет минимальную «хлебную крошку». Очищает состояние сессии.
Управление говорит агенту, что стоит запомнить. Хуки обрабатывают извлечение, дедупликацию, суммаризацию, связывание — чтобы агент не тратил токены на размышления о логистике памяти.
Помимо хуков, mnemos запускает фоновый демон, который постоянно улучшает вашу базу знаний:
Обнаружение устаревания — помечает воспоминания, которые ссылаются на удаленные файлы или устаревшие паттерны
Обнаружение противоречий — находит воспоминания, которые конфликтуют друг с другом
Вывод отношений — автоматически связывает похожие воспоминания
Заполнение данных — ретроактивно генерирует суммаризации для воспоминаний, у которых их нет
mnemos autopilot status # check daemon state
mnemos autopilot run # trigger immediate run
mnemos autopilot run --dry-run # preview findings without writing
mnemos autopilot report # view latest findingsОперация | 350 воспоминаний | 1,500 воспоминаний |
| 57 мс | 24 мс |
| 55 мс | 22 мс |
| 42 мс | 39 мс |
| 27 мс | 108 мс |
hook session-start (холодный) | < 200 мс | — |
размер бинарного файла | ~12 МБ | — |
Оборудование: M1 Pro, 16 ГБ ОЗУ, SQLite на SSD. Ваша задержка может отличаться.
Большинство операций занимают менее 60 мс независимо от размера набора данных. Подкоманды хуков используют режим InitLight — без фоновых воркеров, без прерывания сессии.
Бенчмарк ценности (экономия токенов, точность, избегание ошибок) в процессе. См. DOGFOODING_RUNBOOK.md для методологии. Реальные цифры заменят этот плейсхолдер перед публичным запуском.
Инструмент | Что он делает |
| Сохранить воспоминание (полный авто-конвейер запускается прозрачно) |
| Гибридный FTS + семантический + поиск по пересечению файлов с MMR |
| Собрать диверсифицированный контекст с учетом бюджета для начала сессии |
| Получить по ID |
| Обновить контент, суммаризацию или теги |
| Мягкое удаление (восстанавливаемо через maintain) |
| Связать два воспоминания (supersedes, caused_by, depends_on) |
| Запустить затухание, архивацию, GC, обнаружение устаревших данных |
# Agents call these automatically via MCP. You can also use directly:
mnemos store "JWT uses RS256, 1h expiry, config in auth/config.go"
mnemos search "token expiry"
mnemos stats
mnemos maintainБольшинство пользователей никогда не меняют это. Но если вы хотите:
# ~/.mnemos/config.yaml
embeddings:
provider: noop # noop (default) | ollama | openai
# Pure FTS works fine. Enable semantic for meaning-based search.
quality_gate:
min_words: 5
max_words: 200
min_density: 0.3
require_specific: true # long_term memories need project identifiers
duplicate_threshold: 0.8
summarization:
extractive: true # always on, fast, offline
file_linking:
enabled: true # auto-disables outside git
hook:
enabled: true
search_cooldown: 5m
session_start_max_tokens: 2000
mmr_lambda: 0.7 # 0=max diversity, 1=max relevance
file_boost: 0.3
autopilot:
enabled: true
interval: 15m
contradiction_enabled: falseMnemos классифицирует автоматически. Переопределите вручную через флаг --type.
Тип | Скорость затухания | Использовать для |
| быстрая (~1 день) | задачи, временные заметки, WIP |
| средняя (~1 месяц) | события сессии, исправления багов |
| медленная (~6 месяцев) | архитектурные решения |
| очень медленная | факты, определения, знания |
| быстрая | контекст активной задачи |
Я устал каждое утро заново объяснять свой проект Claude Code.
Я пробовал существующие серверы памяти. Большинство из них отлично хранили текст. Но каждый из них ожидал, что я — или тщательно настроенный промпт — решу, когда сохранять и когда извлекать. Это не база знаний. Это база данных с оберткой MCP.
Mnemos — это то, что я создал, чтобы сделать это действительно автоматическим. mnemos setup claude, перезапустите редактор, и база знаний скомпилирует себя сама.
mnemos setup claude # writes CLAUDE.md, .claude/hooks.json, .mcp.json
mnemos setup cursor # writes .cursorrules, .mcp.json
mnemos setup kiro # writes .kiro/steering/mnemos.md, .kiro/mcp.json
mnemos setup gemini-cli # writes GEMINI.md, .gemini/settings.json, .mcp.jsonФлаги: --global (установка для всех проектов), --force (перезаписать существующее).
mnemos init # first-time setup
mnemos store "..." # store (auto-pipeline)
mnemos search "auth" # hybrid search
mnemos list --project myapp # list memories
mnemos get <id> # fetch by id
mnemos update <id> --content "..." # update
mnemos delete <id> # soft delete
mnemos relate <src> <tgt> --type supersedes # typed relation
mnemos stats # storage + quality stats
mnemos maintain # decay + stale + GC
mnemos serve # MCP server (stdio)
mnemos version
# Autopilot setup
mnemos setup claude | cursor | kiro | gemini-cli [--global] [--force]
# Passive autopilot daemon
mnemos autopilot status
mnemos autopilot run [--dry-run] [--project <id>]
mnemos autopilot report [--project <id>]
# Backfill
mnemos backfill summaries --project <id> [--dry-run] [--limit N]
# Hook subcommands (called by clients, not manually)
mnemos hook session-start
mnemos hook prompt-submit
mnemos hook session-endНе SaaS для памяти чат-бота. Для запоминания предпочтений пользователя в ботах поддержки клиентов используйте Mem0.
Не база данных временных графов знаний. Для рассуждений о бизнес-фактах с учетом времени действия используйте Zep.
Не облачный продукт. Облака mnemos не существует. И никогда не будет.
Не привязан к фреймворку. Нативный MCP. Работает со всем, что поддерживает MCP.
Mnemos делает одно: дает агентам базу знаний, которая компилирует себя сама.
См. ROADMAP.md. Краткая версия:
v1.1.1 (выпущено): полный авто-конвейер, сборка контекста MMR, извлечение с учетом файлов, пассивный демон автопилота, фреймворк бенчмарков
v1.2 (следующее): публичный бенчмарк ценности (dogfooding), npm-обертка, демо-GIF, запуск на HN
v1.3 (планируется): командная память через git — общая .mnemos/shared/ для знаний коллег
v2.0+ (TBD): области памяти между проектами, сжатие памяти, на основе отзывов пользователей
MIT
Distill knowledge into a compiled article
| Name | Required | Description | Default |
|---|---|---|---|
| topic | Yes | Title/subject of the compiled article | |
| content | Yes | The compiled text | |
| project_id | No | Project scope | |
| source_ids | No | Comma-separated source memory IDs |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already indicate destructive behavior, but the description does not explain what gets destroyed or any side effects. It adds no behavioral context beyond the annotations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, concise phrase with no wasted words. However, it may be too brief to be fully informative.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given 4 parameters, no output schema, and annotations indicating destruction, the description is too minimal. It fails to explain how parameters affect the compilation, what the return value is, or side effects.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so the schema already defines parameters. The description does not add any additional meaning or usage hints beyond the field names and types.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The verb 'distill' is specific and the resource 'knowledge into a compiled article' is clear, but it does not differentiate from siblings like mnemos_store or mnemos_update, leaving ambiguity about what 'compile' entails compared to other tools.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides no guidance on when to use this tool versus alternatives, nor does it mention prerequisites or when not to use it. Siblings are listed but not compared.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Assemble relevant context for a query within token budget
| Name | Required | Description | Default |
|---|---|---|---|
| query | Yes | Context query | |
| max_tokens | No | Token budget (default 4000) | |
| project_id | No | Project scope | |
| include_relations | No | Include related memories |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description claims 'assemble context', implying a benign read operation, but annotations indicate destructiveHint=true. The description fails to disclose that this tool may alter or delete state, creating a misleading impression.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single concise sentence that front-loads the primary action. It is appropriately sized for a simple tool, though it could include a bit more context without becoming verbose.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With no output schema and 4 parameters, the description is too sparse. It does not explain what 'context' means, how relations are included, or what the return format is. The behavior around token budget and destructive side effects is not elaborated.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so parameters are already well-documented. The description adds no new semantic information beyond echoing 'query' and 'token budget'. Baseline score of 3 applies.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses the verb 'assemble' and resource 'relevant context', which clearly indicates the tool's purpose is to compile context for a query. It distinguishes from siblings like mnemos_search (which likely searches for specific items) and mnemos_store (which saves), but does not explicitly contrast them.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No guidance on when to use this tool versus alternatives like mnemos_search or mnemos_get. The description does not mention prerequisites or conditions under which this tool is preferred.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Soft-delete a memory
| Name | Required | Description | Default |
|---|---|---|---|
| id | Yes | Memory ID |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The term 'soft-delete' adds value beyond annotations by indicating the operation marks data as deleted without immediate removal. However, no further behavioral details (e.g., reversibility, permission requirements, or state changes) are disclosed.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single phrase with no wasted words. It is front-loaded and efficient for its length.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a one-parameter tool with a soft-delete behavior and no output schema, the description is adequate but lacks depth—e.g., it does not clarify how a soft-deleted memory can be recovered or whether this affects search results.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so 'id' is already documented. The description adds no additional meaning beyond what the schema provides, earning a baseline score of 3.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description 'Soft-delete a memory' uses a specific verb ('soft-delete') and a clear resource ('memory'), making its purpose distinct from sibling tools like mnemos_store or mnemos_get. No ambiguity.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No guidance is provided on when to use soft-delete versus alternatives (e.g., permanent deletion, updates, or other operations). The description lacks context for choosing this tool over siblings.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Get a memory by ID
| Name | Required | Description | Default |
|---|---|---|---|
| id | Yes | Memory ID |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description describes a read operation ('Get'), but annotations set destructiveHint=true, implying mutation or deletion. This is a clear contradiction, and the description does not disclose any behavioral traits beyond what annotations provide.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single concise sentence with no fluff. However, it could be slightly more informative without losing conciseness, hence a 4 rather than 5.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
No output schema is provided, and the description does not explain the return format or structure of the memory object. For a retrieval tool, this is a significant omission, leaving the agent uncertain about what data will be returned.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100% for the single required parameter 'id', with description 'Memory ID' already present in the schema. The description adds no extra semantic meaning beyond the schema, meeting the baseline but not exceeding it.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description 'Get a memory by ID' clearly states the verb (Get), resource (memory), and method (by ID). This distinctly separates it from sibling tools like mnemos_delete or mnemos_store, serving as a straightforward retrieval operation.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No guidance provided on when to use this tool versus alternatives (e.g., mnemos_search). The description lacks any context about prerequisites, exclusions, or specific use cases.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Run decay, archival, and GC maintenance
| Name | Required | Description | Default |
|---|---|---|---|
| project_id | No | Project scope (empty = all) |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The annotations already indicate destructiveHint=true, so the description adds some context by naming the specific maintenance operations (decay, archival, GC). However, it does not disclose what gets destroyed, whether changes are reversible, or other behavioral implications beyond the annotation.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, front-loaded sentence with no wasted words. Every part contributes to conveying the tool's purpose.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The description covers the basic purpose but lacks details about side effects, return values, or how the parameter affects execution. Given the absence of an output schema and the destructive nature, more context would be beneficial for safe and effective use.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100% for the single parameter, so baseline is 3. The description does not add any additional meaning to the parameter beyond what the schema already provides.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the verb 'Run' and specifies the resources 'decay, archival, and GC maintenance', which distinguishes it from sibling tools like mnemos_delete or mnemos_store. However, it could be more specific about what each maintenance operation entails.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides no guidance on when to use this tool versus alternatives, such as when to run maintenance instead of using mnemos_delete or mnemos_update. No context about prerequisites or typical scenarios is given.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Create a relation between two memories
| Name | Required | Description | Default |
|---|---|---|---|
| strength | No | Relation strength [0.0, 1.0] | |
| source_id | Yes | Source memory ID | |
| target_id | Yes | Target memory ID | |
| relation_type | Yes | Relation type: relates_to|depends_on|contradicts|supersedes|derived_from|part_of|caused_by |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations indicate destructiveHint=true and readOnlyHint=false, but the description does not explain whether creating a relation overwrites existing ones or has side effects. The description adds minimal behavioral context beyond what annotations already provide.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single sentence with no wasted words, delivering the core purpose efficiently.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Despite having 4 parameters and no output schema, the description is very brief. It does not explain the effect of the relation, uniqueness constraints, or behavior on duplicates, leaving the agent with significant gaps.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so the schema fully documents all parameters. The description adds no additional meaning to the parameters, meeting the baseline but not exceeding it.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool creates a relation between two memories, which distinguishes it from siblings like mnemos_store (store a memory) or mnemos_delete (delete). However, it could be more specific about what a relation entails in the memory graph.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No guidance is provided on when to use this tool versus alternatives, such as mnemos_context for contextual links. The description lacks context on prerequisites or exclusions.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Report the live MCP server runtime identity: version, host, pid, executable, uptime, data dir, and project scope
| Name | Required | Description | Default |
|---|---|---|---|
No parameters | |||
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Description indicates read-only behavior ('Report'), but annotations set destructiveHint: true, a direct contradiction. The description fails to disclose any actual destructive behavior.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Single sentence, front-loaded with the action and resource, no wasted words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Lists all expected return fields, adequate for a simple info tool. No output schema, so description covers main content, though format or example could strengthen it.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
No parameters, so baseline 4. Description adds context about what the tool reports (version, host, etc.), compensating for the lack of parameters.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
Description clearly states it reports runtime identity and lists specific items (version, host, pid, etc.). Distinct from sibling tools like mnemos_compile, mnemos_delete, etc., which have different purposes.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
Implied usage as a diagnostic/info tool, but no explicit guidance on when to use it vs alternatives. No exclusions or prerequisites mentioned.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Search memories using hybrid text+semantic search
| Name | Required | Description | Default |
|---|---|---|---|
| mode | No | Search mode: text|semantic|hybrid (default hybrid) | |
| limit | No | Max results (default 10) | |
| query | Yes | Search query | |
| project_id | No | Filter by project |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description describes a read-only search operation, but annotations set readOnlyHint=false and destructiveHint=true, creating a contradiction. The description fails to clarify the actual behavioral traits.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single, concise sentence that immediately states the purpose, with no wasted words.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a search tool with no output schema, the description is brief and lacks details about return format, pagination, or behavior, though it covers the core functionality.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 100%, so the description adds no extra meaning beyond the schema. Baseline of 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool searches memories and specifies the method (hybrid text+semantic search), distinguishing it from siblings like mnemos_get or mnemos_delete.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies usage for searching but does not explicitly compare to siblings like mnemos_get or mnemos_context, nor does it provide when-not-to-use guidance.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Store a new memory in Mnemos
| Name | Required | Description | Default |
|---|---|---|---|
| tags | No | Comma-separated tags | |
| type | No | Memory type: short_term|long_term|episodic|semantic|skill|compiled | |
| source | No | Source identifier | |
| content | Yes | Memory content (1 byte to 100KB) | |
| summary | No | Optional summary | |
| category | No | Memory category | |
| project_id | No | Project scope |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description says 'Store a new memory' but annotations indicate destructiveHint=true, implying potential data destruction. The description does not address this contradiction or provide any side-effect context, such as overwriting behavior or resource implications.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
At 6 words, it is very concise but may be too brief for a tool with 7 parameters and no output schema. It front-loads the purpose but lacks supporting detail, making it adequately concise but not optimally informative.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given the complexity (7 params, no output schema), the description fails to mention return values, confirmation behavior, or the consequences of the 'destructive' annotation. It is incomplete for an agent to use effectively without additional context.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
All 7 parameters have schema descriptions (100% coverage), so the description adds no additional meaning. It does not elaborate on how parameters like 'type' or 'tags' affect storage behavior beyond the schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the verb 'store' and resource 'memory' in 'Mnemos', indicating a creation operation. However, it does not differentiate from sibling tools like mnemos_update or mnemos_relate, leaving ambiguity about when to use this tool over others.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No guidance is provided on when to use this tool versus alternatives. There is no mention of prerequisites, exclusions, or context for choosing store over related tools like compile or update.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Update a memory (PATCH semantics)
| Name | Required | Description | Default |
|---|---|---|---|
| id | Yes | Memory ID | |
| tags | No | New comma-separated tags | |
| content | No | New content | |
| summary | No | New summary |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already indicate destructiveHint=true, which aligns with 'Update'. The description adds 'PATCH semantics' but no extra behavioral details (e.g., what happens to unspecified fields, authorization needs). It is minimally adequate.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
A single sentence that is perfectly concise and front-loaded. Every word contributes essential information (action, resource, update semantics).
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a destructive tool with no output schema, the description lacks context about return values, side effects, or behavior on error. More detail would be needed for safe invocation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Input schema covers all 4 parameters with descriptions (100% coverage). The tool description adds no additional meaning beyond the schema, so baseline score of 3 is appropriate.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description 'Update a memory (PATCH semantics)' clearly states the action (update) and the resource (memory). The parenthetical note adds specificity about partial updates, distinguishing it from siblings like mnemos_store (create) or mnemos_delete.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No guidance on when to use this tool vs alternatives like mnemos_get or mnemos_search. No prerequisites or exclusions are mentioned, leaving the agent without decision context.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v1.2.1mnemos_runtimev1.1.14mnemos_compilemnemos_store1 field changedInput schema / properties / type / descriptionPrevious value: -"Memory type: short_term|long_term|episodic|semantic"New value: +"Memory type: short_term|long_term|episodic|semantic|skill|compiled"v0.1.2mnemos_relate1 field changedInput schema / properties / relation_type / descriptionPrevious value: -"Relation type"New value: +"Relation type: relates_to|depends_on|contradicts|supersedes|derived_from|part_of|caused_by"v0.1.0mnemos_contextmnemos_deletemnemos_getmnemos_maintainmnemos_relatemnemos_searchmnemos_storemnemos_updateScored across 10 tools
Each tool has a clearly distinct purpose: CRUD operations (store, get, update, delete), search, maintenance, relation creation, context assembly, compilation, and runtime info. No two tools overlap in functionality, ensuring an agent can easily select the correct tool.
All tools follow a consistent 'mnemos_' prefix with an underscore-separated verb or noun. Most use imperative verbs (compile, delete, get, maintain, relate, search, store, update), while 'context' and 'runtime' are nouns. This minor inconsistency prevents a perfect score.
With 10 tools, the surface is well-scoped for a memory/knowledge server. It covers essential CRUD, search, maintenance, relations, and advanced features like compilation and context assembly without being overwhelming or sparse.
The tool set covers the full memory lifecycle (create, read, update, soft-delete) plus advanced operations (compile, context, relate, maintain, runtime). Minor gaps include missing batch operations or explicit undo for soft-delete, but the core domain is well-served.
Persistent memory for AI agents. Semantic search, memory graph, W3C DID identity.
Persistent memory for AI agents. Search and store durable facts, preferences and decisions.
Persistent, inspectable memory for AI agents with lineage, correction, and a hosted MCP endpoint.
Persistent memory for AI agents across Claude, ChatGPT and any MCP client.
