Answer42
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Answer42 open the "GoodsReceipt" form and click "Post""
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Answer42

Answer42 — MCP-инструмент для интерактивного управления UI 1С:Предприятия через клиент тестирования. Он позволяет AI-агенту открывать формы 1С, нажимать кнопки, заполнять поля, выбирать ссылки из форм выбора, работать с таблицами, динамическими списками и табличными документами.
The Answer to Life, Universe, and 1C — UI Driver
Проект даёт AI-агенту три способности:
Интерактивное управление UI 1С — открыть форму, кликнуть кнопку, активировать поле, заполнить значение, выбрать строку, провести сценарий через
/TESTMANAGER+/TESTCLIENT.Доказательная запись клиентского тестирования — записывать аннотированные PDF-слайды с MCP method/request/response и скриншотами окна 1С.
RAG-индекс метаданных — локальный SQLite/FTS индекс конфигураций 1С (EDT, XML-выгрузка Конфигуратора, base + extensions), включая подсказки для
ui_treeи dynamic-list settings.
Быстрый старт
# Рекомендуется для обычной установки из PyPI
pipx install 'answer42[screenshot,linux-window-control]'
# Альтернатива без pipx
python3 -m venv .venv
. .venv/bin/activate
pip install 'answer42[screenshot,linux-window-control]'
# Разработка из git checkout
pip install -e '.[screenshot,dev,linux-window-control]'
# Windows: запускать из интерактивной пользовательской сессии, не как service
# pipx install "answer42[screenshot,windows-window-control]"
# pip install "answer42[screenshot,windows-window-control]"
# Запуск MCP-сервера (transport: stdio)
answer42 --ws-host 127.0.0.1 --ws-port 8765Запуск сессии и подключение тестируемой базы выполняются одним MCP-вызовом:
start_session(
session_id="my-session",
base_url="<TARGET_INFOBASE_URL>",
idle_timeout_minutes=60,
execute="/path/to/external.epf", # опционально: /Execute
command_parameter="InitScenario" # опционально: /C
)start_session определяет версию 1С по base_url, поднимает инфраструктуру Answer42 (на Linux — Xvfb, если нужен headless; на Windows — интерактивная desktop-сессия), свободные порты, WebSocket bridge, файловую базу менеджера и test-manager. Если доступны серверные компоненты 1С (ibsrv), менеджер и встроенная тестовая база публикуются через автономный сервер; если доступен только тонкий клиент и ibcmd, Answer42 автоматически использует fallback file-direct (/F) для файловых баз менеджера и клиента. После idle_timeout_minutes минут неактивности сессия автоматически завершается; значение по умолчанию — 60 минут. Для запуска клиента можно передать extra_args, execute (параметр командной строки /Execute) и command_parameter (параметр /C); если /C уже указан в extra_args, значения склеиваются через ;.
Логин/пароль рекомендуется не передавать явно: start_session умеет брать их из локального credentials-файла по base_url. Параметры username/password остаются в API для разовых сценариев и обратной совместимости, но при передаче в MCP-вызове они видны вызывающему агенту в аргументах tool-call.
Related MCP server: bas-mcp-server
ui_tree profiles and output formats
ui_tree is optimized for form-structure diagnostics. The default profile="navigation" returns a compact visible UI tree without data presentations, RAG enrichment, or command-panel subtrees. The tree is not depth-limited.
Useful modes:
ui_tree(parent_name=...)serializes only the subtree under a known parent element, useful for large/complex forms.ui_tree(name=.../title=.../type=...)returns a flatobjectslist instead of a nested tree; on current managers filtering is performed on the 1C side; simple exactname/typequeries may use the platform fast path before falling back to flat traversal.profile="diagnostic"includes command panels without enabling expensive data reads.profile="data"enables node data presentations; prefer targetedfield_value_text,table_rows,tabular_document_text, ortabular_document_saveunless you really need data for many nodes.profile="full"gives the closest historical verbose result.fields="minimal|navigation|diagnostic|data|full"(or a comma-separated custom field list) controls node payload without truncating the tree.command_panels="include|exclude|only"controls command-panel serialization explicitly;onlyreturns a deduplicated flat list.group_mode="include|flatten|flatten_layout|exclude|exclude_empty|pages_only"controls whether group nodes are kept, lifted, or omitted.format="json|outline|yaml|yml"keeps JSON as the default structured result and offers readable text views for inspection. Invalid profile/group/command-panel/format values are rejected; every result includesmetricswith timings/payload counters;tree_summary.types_countreports node counts by broad family (buttons,fields,tables,groups, etc.) and exact 1C test-client type.
Large ui_tree snapshots as resources
By default delivery="auto" keeps small results inline and automatically stores large results as resources when the estimated token count exceeds ONEC_MCP_UI_TREE_RESOURCE_AUTO_THRESHOLD_TOKENS (default 50k).
Use delivery="inline" to force the selected JSON/outline/yaml directly in the response. Use delivery="resource" to force resource storage for large diagnostics:
r = ui_tree(profile="full", include_hidden=true, delivery="resource")When auto switches to resource, or when delivery="resource" is forced, the inline response contains only resource_uri, snapshot_id, summary, metrics, byte size, estimated tokens, threshold reason and a small preview. The full JSON snapshot is stored under the Answer42 runtime data directory and registered as an MCP resource (answer42://ui-tree/<session>/<snapshot_id>). Do not load that whole resource into the agent context unless you really need the raw file; use targeted snapshot tools instead:
ui_tree_resource_search(resource_uri=..., name/title/type/text=...)— find a few controls in a saved snapshot and get theirpath/index_path.ui_tree_resource_get_node(resource_uri=..., index_path=...)— read exactly one node.ui_tree_resource_children(resource_uri=..., index_path=..., depth=1|2)— read a small subtree.ui_tree_diff(before_resource_uri=..., after_resource_uri=...)— compare snapshots before/after a UI action.ui_tree_resource_action(resource_uri=..., index_path=..., action="focus|click_button|activate_field|field_value_text|field_info")— reuse a node selector from the snapshot and execute a first-class live UI tool.
index_path is the preferred stable address inside one snapshot. path is human-readable and useful for reports, but less reliable when several controls have the same captions.
Встроенные рекомендации для агентов
Answer42 больше не поставляет и не устанавливает bundled skills. Рекомендации по UI-автоматизации, RAG, Универсальному отчёту и экспорту табличных документов находятся в descriptions соответствующих MCP-инструментов и встроенных MCP prompts.
Команды answer42 install-skills и answer42 skills-install сохранены как
совместимые no-op: они ничего не копируют и выводят сообщение о новой модели.
Инструкция для AI-агентов
Если вы AI-агент и пользователь просит установить Answer42, используйте отдельную инструкцию: docs/agent-installation.md.
Коротко:
установите пакет через
pipxили virtualenv;подготовьте
ONEC_MCP_CREDENTIALS_FILEвне репозитория;зарегистрируйте MCP-сервер в агентском клиенте;
проверьте установку через
credentials_check, затем smoke-сессиюstart_session→active_window→stop_session.
Linux display prerequisite
На Linux start_session использует живую X11-сессию, если процесс Answer42 унаследовал непустой DISPLAY: test manager и test client запускаются на этом дисплее, поэтому их окна и скриншоты видны на desktop.
Если DISPLAY не задан (headless service, SSH без X11 forwarding и т. п.), Answer42 поднимает два собственных изолированных Xvfb-дисплея: один для test manager и второй для test client. В этом режиме пакет xvfb обязателен:
# Debian / Ubuntu
sudo apt install xvfb
# Fedora / RHEL
sudo dnf install xorg-x11-server-XvfbЕсли не задан ни доступный DISPLAY, ни Xvfb, start_session завершается до создания ресурсов с понятной подсказкой по установке. Чтобы desktop-сервис видел живой X11, его launcher/service должен передать корректный DISPLAY и права доступа к X server (обычно через XAUTHORITY).
Скриншоты: stdio и StreamableHTTP
screenshot сохраняет PNG на MCP-хосте и больше не помещает его base64-представление в результат tool-call.
В stdio возвращаются только
path, размер и диагностические поля. Агентский хост при необходимости прикладывает файл по пути.В StreamableHTTP добавляется непрозрачная одноразовая ссылка
urlна PNG. Она действует один час по умолчанию (ONEC_MCP_SCREENSHOT_URL_TTL_SECONDS), исчезает после первого скачивания и существует не дольше процесса сервера. Для внешнего reverse proxy укажите публичный base URL черезONEC_MCP_SCREENSHOT_URL_BASE.
Транспорты: stdio и StreamableHTTP
По умолчанию Answer42 использует stdio MCP transport (запускается MCP-хостом как subprocess). Для удалённого deployment доступен StreamableHTTP режим с MCP auth:
answer42 --http --http-host 0.0.0.0 --http-port 8080 \
--http-token "static-secret-token" \
--http-account-id "tenant-42"Переменные окружения:
export ONEC_MCP_HTTP=1
export ONEC_MCP_HTTP_HOST=0.0.0.0
export ONEC_MCP_HTTP_PORT=8080
export ONEC_MCP_HTTP_TOKEN="static-secret-token"
export ONEC_MCP_ACCOUNT_ID="tenant-42"
answer42Каждый MCP-запрос должен содержать заголовок
Authorization: Bearer <token>.account_idопределяет namespace для credentials. Если не задан, вычисляется как стабильный хеш токена (token-<16hex>).Режим по умолчанию stateless (
--http-stateless/ONEC_MCP_HTTP_STATELESSdefault1); для resumable sessions передай--http-stateless/0.
Безопасное хранение логинов и паролей
Чтобы креды были доступны MCP-серверу, но не попадали в чат и аргументы tool-call, храните их в локальном файле за пределами репозитория и передайте путь в окружение процесса Answer42:
export ONEC_MCP_CREDENTIALS_FILE=/secure/path/credentials.jsonЕсли переменная окружения не задана, MCP-сервер читает файл по умолчанию:
~/.answer42-credentials.jsonФормат файла v2 (аккаунт-bound):
{
"version": 2,
"accounts": {
"tenant-42": {
"entries": [
{
"url": "https://example.invalid/infobase",
"username": "<USERNAME>",
"password": "<PASSWORD>",
"title": "dev-example",
"aliases": ["dev42"]
},
{
"url": "https://*.example.invalid/*",
"username": "<WILDCARD_USERNAME>",
"password": "<WILDCARD_PASSWORD>",
"title": "wildcard-example",
"aliases": ["we"]
}
]
}
}
}Устаревший v1 формат (flat entries) автоматически загружается в account legacy и мигрирует в v2 при следующем сохранении. Рекомендуется явно задать account_id (через HTTP auth claim или ONEC_MCP_ACCOUNT_ID в stdio) и перенести записи в соответствующий account.
Рекомендуемые права на файл: 0600.
Правила матчинга внутри аккаунта: сначала точное совпадение url, затем wildcard (*, ?) через fnmatch; первое совпадение побеждает. Пароли не логируются и не возвращаются наружу.
Для сохранения или обновления записи можно использовать MCP-tool:
credentials_save(
url="<TARGET_INFOBASE_URL>",
username="<USERNAME>",
password="<PASSWORD>"
)Для удаления записи:
credentials_remove(url="<TARGET_INFOBASE_URL>")Для проверки, что для адреса есть сохранённые креды, используйте MCP-tool:
credentials_check(base_url="<TARGET_INFOBASE_URL>")Ответы этих tools содержат только факт наличия/изменения/удаления записи и URL; логины и пароли не возвращаются. credentials_list() дополнительно показывает статус проверки (verified / verification_status): новые или изменённые записи сохраняются как unverified, после успешного start_session с этой учёткой помечаются как verified, а unverified запись удаляется при ошибке авторизации. Tool credentials_list() оставлен в коде для локальной диагностики, но в OpenClaw-конфигурации его рекомендуется скрывать через toolFilter.exclude, чтобы агент не мог получить список URL-шаблонов.
Остановка сессии:
stop_session(session_id="my-session", clean_data=False)Требования
Python 3.11+
1С:Предприятие 8.3.27+ или 8.5+
Пакеты Python:
mcp,websockets,pydantic; для скриншотов —mssLinux/X11:
python-xlib/wmctrl/xdotool, GUI/Xvfb для headless-сервера; OS-слой допускается для подготовки геометрии окна (resize/maximize), но не для ввода/кликов/автоматизации 1СWindows: интерактивная пользовательская desktop-сессия; window-control и screenshots работают через WinAPI/
mss, ввод/клики/автоматизация 1С выполняются только через API клиента тестирования
Проверяйте наличие платформы 1С в стандартных каталогах: Linux /opt/1cv8/x86_64/<version>/ и /opt/1cv8/i386/<version>/; Windows C:\Program Files\1cv8\<version>\bin\ и C:\Program Files (x86)\1cv8\<version>\bin\; macOS /Applications/1cv8/<version>/ или /opt/1cv8/<version>/. Для штатной работы нужны 1cv8c и ibcmd; ibsrv желателен, но при его отсутствии Answer42 может использовать fallback /F. Если автоопределение ошиблось, задайте ONEC_PLATFORM_DIR. Для очень медленного старта web-клиента можно увеличить ONEC_MCP_TEST_CLIENT_READY_TIMEOUT и timeout MCP-клиента; по умолчанию Answer42 ждёт открытия -TPort 55 секунд и затем отдаёт явную ошибку с логами клиента.
Безопасность стендов и учётных данных
В репозитории не должно быть реальных URL стендов, логинов или паролей. Для штатного запуска передавайте только base_url, а логин/пароль храните в локальном credentials-файле, доступном процессу Answer42 через ONEC_MCP_CREDENTIALS_FILE.
start_session всё ещё принимает username и password напрямую для разовых сценариев и обратной совместимости. Используйте это только когда осознанно готовы раскрыть значения вызывающему агенту: параметры MCP-вызова могут попасть в историю чата, логи клиента или отладочный вывод. Если вместо реального пароля передан редактированный плейсхолдер из звёздочек (***, ******** и т.п.), Answer42 остановит запуск с явной ошибкой: нужно указать настоящий пароль или сохранить корректные креды через credentials_save().
Примеры в документации используют только плейсхолдеры (<TARGET_INFOBASE_URL>, <USERNAME>, <PASSWORD>). Перед публикацией артефактов проверяйте, что параметры вызовов заредактированы: recorder маскирует ключи вроде password, но URL и логин тоже не должны попадать в публичные материалы.
Архитектура
Поток выполнения:
MCP-клиент вызывает tools через stdio MCP.
Answer42 принимает MCP-вызовы и передаёт команды в WebSocket bridge.
1С test manager подключается к bridge и выполняет BSL-dispatch.
Test manager управляет 1С test client через API
ТестируемоеПриложение.Test client выполняет интерактивные операции в целевой информационной базе.
Подробнее: docs/architecture.md.
Сборка конфигураций 1С
Конфигурация менеджера тестирования собирается из XML-исходников src/cf/ через временную файловую ИБ: ibcmd infobase config import и затем ibcmd config save (приоритетный способ). Такой путь создаёт переносимый .cf с compatibility mode XML-исходников; Конфигуратор/DESIGNER используется как fallback, если ibcmd отсутствует:
python scripts/build_cf.py src/cf build/MCPTestManager.cfPowerShell:
python scripts/build_cf.py src/cf build/MCPTestManager.cfТестовая конфигурация для E2E собирается из XML-исходников src/client_cf/:
python scripts/build_cf.py src/client_cf build/MCPTestClient.cfPowerShell:
python scripts/build_cf.py src/client_cf build/MCPTestClient.cfВ Git хранятся только XML-исходники; CF — генерируемые артефакты. start_session использует build/MCPTestManager.cf только при точном совпадении content fingerprint с src/cf/; иначе автоматически пересобирает его локальной выбранной платформой. Встроенный demo-клиент аналогично работает с build/MCPTestClient.cf и src/client_cf/. Релизный wheel получает оба CF из GitLab CI, где они собираются закреплённой платформой 1С 8.3.27.2342.
E2E можно запускать целиком или по независимым сценариям:
python3 scripts/e2e_stable.py # full
E2E_SCENARIO=smoke python3 scripts/e2e_stable.py # быстрый smoke
E2E_SCENARIO=dynamic python3 scripts/e2e_stable.py # legacy: таблицы/dynamic-list/отчёт
E2E_SCENARIO=dynamic-tables python3 scripts/e2e_stable.py
E2E_SCENARIO=dynamic-lists python3 scripts/e2e_stable.py
E2E_SCENARIO=dynamic-reports python3 scripts/e2e_stable.py
E2E_SCENARIO=coverage python3 scripts/e2e_stable.py # diagnostic/negative toolsДля реального распараллеливания сценарий умеет сам запустить пять разные сессий (smoke, dynamic-tables, dynamic-lists, dynamic-reports, coverage), а не копии одного и того же теста. Встроенная тестовая файловая база публикуется через один общий ibsrv, когда серверные компоненты доступны; без ibsrv demo-клиент запускается через file-direct (/F):
E2E_PARALLEL=1 E2E_SESSION_ID=e2e-split python3 scripts/e2e_stable.pystart_session также переиспользует общий автономный сервер для одинаковой файловой базы. Последний stop_session освобождает refcount и завершает shared ibsrv.
Ограничения
Форма пользовательской настройки «Изменить форму» частично недоступна для надёжной автоматизации через API клиента тестирования. В частности, команда «Добавить поля» в верхней панели формы настройки может быть видна на скриншоте и в диагностическом тексте, но не нажиматься как обычная
ТестируемаяКнопкаФормы:click_buttonможет не находить её, аfocus_objectможет вернуть успешную фокусировку без открытия диалога добавления полей.
Лицензия
MIT
Copyright
Copyright (c) 2026 Kosolapov Stanislav aka proDOOMman prodoomman@gmail.com, Marvin (AI Assistant), 42Clouds, and contributors.
Licensed under the MIT License.
Available Tools
129 toolsactivate_fieldC
Activate/focus a form field by name or title.
Inputs: field_name (required), session_id, search_by_title.
activate_windowB
Activate/switch to a tested 1C client window by index, title or name.
Inputs: name, index, title, session_id.
active_windowC
Return active 1C test client window state.
Inputs: session_id.
choose_current_rowC
Choose the current row in the active selection/list form.
Inputs: name, session_id, expected_text.
choose_field_first_rowC
Open reference selection form and press its standard Choose button.
Inputs: field_name (required), session_id, search_by_title.
choose_field_from_listC
Open the standard selection form for a reference field.
Inputs: field_name (required), session_id, search_by_title.
clear_file_dialog_resultC
Clear queued file selection dialog results.
Inputs: session_id.
click_buttonC
Find a button in the active UI tree and press it.
Inputs: name (required), session_id, search_by_title, watchdog_timeout.
client_busy_statusC
Check whether the 1C test client/bridge is responsive and whether a dangerous action is blocking the interface.
Inputs: session_id, probe_timeout.
close_formC
Close the active form.
Inputs: session_id.
close_user_messages_panelB
Close the user messages panel in the active test-client window.
Inputs: session_id.
command_barA
Low-level diagnostic: return active window/form command-bar data when available.
Prefer command_interface_sections/commands/search for application navigation commands, and object_command_panel/object_context_menu for commands of a specific UI object.
Inputs: session_id, include_timings.
command_interface_commandsB
Return command-interface command tree and flattened command list, optionally filtered by section.
Inputs: section, session_id, window_title.
command_interface_searchC
Search live command-interface commands by title, path or navigation URL.
Inputs: query (required), section, session_id, max_results, +2 more.
command_interface_sectionsC
Return top-level sections of the active window command interface.
Inputs: session_id, window_title.
configuration_name_version_from_subsystemsC
Open the 'Subsystem versions' information register and return the main configuration name/version.
The tool reads РегистрСведений.ВерсииПодсистем via the list output
command so that columns available to the dynamic list, including
Это основная конфигурация, are visible even when they are not displayed
in the default form table.
Inputs: session_id, refresh_cache. Returns: object
credentials_checkB
Check whether credentials exist for a base URL or saved title without returning secrets.
Inputs: base_url (required). Returns: object
credentials_listA
List configured credential URL patterns/titles (no usernames or passwords).
Inputs: no named parameters. Returns: result
credentials_removeA
Remove a credential entry from the local store by exact URL match.
Inputs: url (required). Returns: object
credentials_saveA
Save or update credentials for a URL pattern in the local credentials store.
The credentials file is never committed to git. Use this tool to persist
login/password pairs so that start_session can pick them up automatically
without exposing secrets in MCP tool-call arguments. Entries are scoped to the
current MCP account.
Inputs: url (required), password (required), username (required), title, +1 more. Returns: object
credentials_titlesA
List saved credential titles/aliases that can be used as start_session base_url.
Inputs: no named parameters. Returns: result
current_elementC
Return the current active form element.
Inputs: session_id.
current_error_infoC
Return current 1C script error information displayed in the tested client, if any.
Inputs: session_id.
decoration_clickD
Click a form decoration.
Inputs: name, title, session_id.
decoration_click_linkC
Click a hyperlink inside a formatted-string decoration.
Inputs: name, title, session_id, presentation.
decoration_linksC
Return hyperlink presentations from a formatted-string decoration.
Inputs: name, title, session_id.
dynamic_list_clear_settingsC
Quickly reset/clear dynamic-list filter/order/conditional formatting/grouping settings.
Inputs: button, session_id.
dynamic_list_current_settingsA
Open the dynamic-list settings form and read currently configured filter/order/etc. items.
This reads the selected section's configured elements table, not the 'Доступные поля' table. Use dynamic_list_settings_available_fields when you need fields that can be added. Read current settings before changing them; do not clear or replace existing filters/orders unless the requested scenario explicitly requires it.
Inputs: close, section, session_id.
dynamic_list_findC
Search a dynamic list through context menu 'Расширенный поиск' and the standard modal 'Найти' dialog.
Inputs: text (required), name, field, title, +2 more.
dynamic_list_metadata_available_fieldsA
Return RAG/metadata-known fields for a metadata object; this does not inspect the live UI settings form.
Inputs: object, session_id, rag_snapshot.
dynamic_list_open_settingsC
Open dynamic-list settings (filter/order/conditional formatting/grouping) via 'Настроить список'.
Inputs: button, session_id.
dynamic_list_outputC
Press dynamic list 'Вывести список' button and return detected tabular document(s), optionally including text.
Inputs: button, max_rows, session_id, max_columns, +2 more.
dynamic_list_set_settingsA
Open the dynamic-list settings form and add filter/order elements through the standard dialog.
Filters: [{field, value, comparison?}] — comparison defaults to 'Равно'. Orders: [{field, direction?}] — direction defaults to empty (uses dialog default).
This is the preferred way to apply settings; it uses the same dialog the user sees.
Keep clear_first=false unless replacing all user settings was explicitly requested.
Inputs: close, orders, filters, session_id, +1 more.
dynamic_list_settings_available_fieldsA
Open the dynamic-list settings form, switch to the requested section tab, and read the 'Доступные поля' table.
Returns the live list of fields available for filtering/ordering/conditional-formatting/grouping in the current dynamic list. This is the authoritative source — it reflects the actual form, not RAG metadata.
Inputs: close, limit, section, session_id, +1 more.
dynamic_list_sort_by_columnA
Quickly sort the active dynamic list by a visible column title via the 1C test-client API УстановитьПорядок; this is not the settings-form Sort tab.
Inputs: column_title (required), name, title, session_id.
export_eventlogC
Export the 1C event log.
Inputs: fmt, to_time, out_path, from_time, +3 more. Returns: object
field_dropdown_closeB
Close an open dropdown for a form field, if it is open.
Inputs: field_name (required), session_id, search_by_title.
field_dropdown_itemsA
Open a form field dropdown and return visible choice items when exposed by the 1C test-client API.
This is a convenience action for the common "open dropdown and read choices" scenario. For non-mutating diagnostics of an already-open dropdown use field_dropdown_state.
Inputs: field_name (required), timeout, session_id, search_by_title.
field_dropdown_stateC
Return whether a field dropdown is open and its visible choice items.
Inputs: field_name (required), timeout, session_id, search_by_title.
field_infoB
Return diagnostics for a form field: visibility, availability, readonly, kind, title.
Inputs: field_name (required), session_id, search_by_title.
field_tooltipC
Return field tooltip text.
Inputs: field_name (required), session_id, search_by_title.
field_value_textC
Return field data presentation text via ПолучитьПредставлениеДанных().
Inputs: field_name (required), session_id, search_by_title.
find_objectC
Find an object in the active UI hierarchy by name, title, or test object type.
Inputs: name, type, title, session_id.
focus_objectA
Move focus to a UI object in the active test client window.
This tool only changes focus/activation; it does not change values. For radio buttons and dropdowns use set_field_value with the visible caption. To switch to an inactive page, focus the page object first, then interact with its children.
Inputs: name, type, title, session_id.
form_default_buttonC
Return the active form default button, when available.
Inputs: session_id.
form_next_elementA
Move focus to the next form element and return the current element.
Inputs: session_id.
form_open_settingsC
Open the current form customization dialog via 'Изменить форму'.
Inputs: button, session_id.
form_previous_elementC
Move focus to the previous form element and return the current element.
Inputs: session_id.
form_stateC
Return active form availability/modified/readonly state.
Inputs: session_id.
get_objectA
Low-level escape hatch: call TestedApplication.GetObject with optional wildcard name/title.
Prefer find_object/search_objects for structured diagnostics and focus_object for changing focus. Keep this tool for platform/test-client edge cases where direct GetObject semantics matter.
Inputs: name, type, title, timeout, +1 more.
goto_next_windowB
Navigate the main test-client window to the next open window.
Inputs: session_id.
goto_previous_windowC
Navigate the main test-client window to the previous open window.
Inputs: session_id.
goto_start_pageC
Navigate the main test-client window to the start page.
Inputs: session_id.
group_activate_pageC
Activate/switch to another page in a page-group form element.
Inputs: name, title, page_name, page_title, +1 more.
group_collapseC
Collapse a collapsible form group.
Inputs: name, title, session_id.
group_current_pageC
Return current page of a page-group form element.
Inputs: name, title, session_id.
group_expandC
Expand a collapsible form group.
Inputs: name, title, session_id.
group_stateC
Return form group expanded/visible/available state.
Inputs: name, title, session_id.
installed_extensions_from_catalogB
Open Справочник.ВерсииРасширений and return installed extensions from the latest non-deleted metadata snapshot.
The tool reads catalog rows, ignores rows marked for deletion, takes the latest
row by last-use date/code and parses Описание метаданных. Metadata lines
starting with EF_ and the main configuration line starting with # are not
returned as extensions.
Inputs: session_id, refresh_cache. Returns: object
metadata_objects_catalog_cache_clearB
Clear cached metadata objects for the current session.
Inputs: session_id. Returns: object
metadata_objects_from_catalogA
Return/search metadata objects from Справочник.ИдентификаторыОбъектовМетаданных.
The first call opens the standard Universal report for the metadata-identifiers
catalog, reads a normalized snapshot with metadata names/synonyms, excludes
deleted rows plus roles/subsystems, and stores the result in an in-memory
per-session cache. Later calls search that cache unless refresh_cache=true.
This is a live UI fallback when RAG is unavailable or stale; if its watchdog
reports a blocked interface, wait for client recovery before further UI calls.
Inputs: limit, query, session_id, max_results, +3 more. Returns: object
_metadata_objects_from_catalog_implD
Inputs: limit (required), query (required), session_id (required), max_results (required), +2 more. Returns: object
object_command_panelB
Return command panel for a found UI object, when supported.
Inputs: name, type, title, session_id.
rag_available_configurationsB
List RAG-available configurations, extensions, snapshots and discovery roots.
Inputs: sync, session_id. Returns: object
rag_index_buildC
Build/rebuild the SQLite/FTS RAG index for a source or snapshot.
Inputs: source, snapshot, session_id. Returns: object
rag_lookup_objectA
Look up an indexed 1C metadata object, including navigation, fields and forms.
RAG describes configuration source metadata, not live infobase data,
access rights, user settings, or a report result. Inspect rag_diagnostic
before concluding that an object is absent, then use live UI tools when
current runtime state matters.
Inputs: object (required), snapshot, session_id. Returns: object
rag_queryA
Search business terms and full-text RAG chunks in the local SQLite index.
Use RAG to discover metadata names, fields, forms, and base/extension
layers; it is not evidence of current live data, permissions, or report
output. Inspect rag_diagnostic if search returns no useful result.
Inputs: query (required), limit, snapshot, session_id. Returns: object
rag_session_autoconfigureA
Auto-configure the session RAG snapshot from the live 1C configuration and installed extensions.
The tool reads the live main configuration/version from ВерсииПодсистем, reads
installed extensions from ВерсииРасширений, matches them against registered
RAG sources by metadata name and version, and creates/binds a snapshot.
When no exact/suitable source exists and dump_if_missing=true, Answer42 can
create RAG sources by dumping the configuration and extensions to Designer XML.
Dumping requires a direct Designer connection: server infobase (server/ib) or
file infobase path. Web /WS URLs cannot be dumped; for web sessions pass
dump_connection_value, file_infobase_path, or cf_files/cfe_files.
If the direct dump connection has different users than the web publication
(for example multitenant/separated-data infobases), pass dump_username and
dump_password; otherwise the current session credentials are reused.
Temporary infobases created from CF/CFE files are deleted after dumping by
default. Generated XML source directories are removed after a successful
index build by default, because the SQLite RAG index is self-contained; pass
cleanup_generated_sources=false when the dumped XML source tree should be
kept under dump_root for inspection or reuse.
Inputs: cf_files, cfe_files, dump_root, session_id, +12 more. Returns: object
rag_snapshot_createC
Create/update an effective configuration snapshot from base + extensions.
Inputs: name (required), base, sources, extensions, +1 more. Returns: object
rag_snapshots_listC
List effective RAG snapshots and their layers.
Inputs: session_id. Returns: result
rag_source_addC
Register/update a 1C source tree; format is detected automatically when format=auto.
Inputs: name (required), path (required), kind, format, +1 more. Returns: object
rag_source_roots_listC
List RAG source discovery roots and optionally sync newly added configurations/extensions.
Inputs: sync, session_id. Returns: object
rag_sources_add_from_directoryC
Discover/register 1C configuration and extension source directories under a root directory.
Inputs: path (required), watch, prefix, recursive, +1 more. Returns: object
rag_sources_listC
List registered 1C source trees.
Inputs: sync, session_id. Returns: result
recording_captureC
Manually capture a screenshot/slide in the active recording session.
Inputs: action (required), note, session_id. Returns: object
recording_startB
Start automatic evidence recording: each recorded tool call gets screenshot + annotated slide.
Inputs: window, display, output_dir, session_id. Returns: object
recording_stopA
Stop recording and build PDF evidence from captured slides.
Inputs: session_id. Returns: object
save_formC
Save/write the active form.
Inputs: session_id.
screenshotA
Capture a screenshot from the MCP host. Defaults to the active visible 1C window.
The response always contains the saved path, never image base64. In
StreamableHTTP mode it additionally contains an opaque download URL that
expires after one hour by default (ONEC_MCP_SCREENSHOT_URL_TTL_SECONDS).
Screenshots are for user evidence and UI diagnostics only. Do not use them
to analyze report/table/tabular-document/list data; use structured tools
such as tabular_document_text, tabular_documents, table_rows,
dynamic_list_output, or report export/save tools instead.
Inputs: path, window, display, session_id. Returns: object
search_objectsC
Find all UI objects matching name, title and/or 1C test object type.
Inputs: name, type, limit, title, +1 more.
sessions_listA
List active/restorable Answer42 sessions and their connection state.
Inputs: no named parameters. Returns: object
session_statusC
Return bridge, runtime, test-client and RAG state for a session.
Inputs: session_id. Returns: object
set_field_valueC
Set a value in a form field by name or title.
Inputs: value (required), field_name (required), session_id, search_by_title, +1 more.
set_file_dialog_resultC
Preload the result for the next 1C file selection dialog.
Inputs: paths (required), session_id.
set_form_fieldsC
Set multiple ordinary form fields using the same logic as set_field_value.
Inputs: values (required), session_id, search_by_title, create_if_missing.
start_sessionA
One-shot session factory: detect version, start infrastructure, and connect the 1C test client.
base_url may be a web publication URL, server/infobase, or a file
infobase path. Prefer credentials saved with credentials_save over
passing secrets in tool calls. After a long UI operation, use
client_busy_status or wait_client_idle before sending more commands.
Inputs: execute, version, base_url, password, +8 more. Returns: object
stop_sessionB
Tear down a session: disconnect test-client, stop test-manager, ibsrv, Xvfb and bridge.
Inputs: clean_data, session_id. Returns: object
table_add_rowC
Add a row to the active tested table via the 1C test-client API (Insert / plus).
Inputs: name, title, session_id.
table_begin_editB
Begin editing the current row in a form table.
Inputs: name, title, session_id.
table_cell_textC
Return cell text from a tested table via GetCellText().
Inputs: cell, name, title, session_id.
table_choose_field_from_listC
Open the selection form for a reference field in a form table row.
Inputs: field (required), table (required), session_id, search_by_title.
table_clear_selectionC
Clear all row selection in a form table.
Inputs: name, title, session_id.
table_copy_rowC
Copy the current row in a form table.
Inputs: name, title, session_id.
table_current_rowC
Return current row of a tested table as key/value text mapping.
Inputs: name, title, session_id.
table_delete_rowC
Delete the current row in a form table.
Inputs: name, title, session_id.
table_edit_infoC
Return diagnostics for currently edited table row and its child fields.
Inputs: table, session_id.
table_editing_stateC
Return whether a form table is currently in edit mode.
Inputs: name, title, session_id.
table_end_editC
End editing the current row in a form table.
Inputs: name, title, session_id.
table_find_rowC
Find a tested table row containing text.
Inputs: text (required), name, limit, title, +1 more.
table_goto_rowC
Move cursor to the Nth tested table row containing text.
Inputs: text (required), name, limit, title, +2 more.
table_move_rowD
Move current row using official test-client table navigation methods.
Inputs: name, steps, title, direction, +1 more.
table_rowsC
Return rows of a tested table as key/value mappings.
Inputs: name, limit, title, compact, +1 more.
table_select_allC
Select all rows in a form table.
Inputs: name, title, session_id.
table_selected_rowsC
Return selected rows of a tested table as key/value text mappings.
Inputs: name, title, session_id.
table_set_fieldA
Set a field in the currently edited row of a form table.
Finds the editor field for the column (e.g. ТоварыНоменклатура), activates it, delegates to the general set_field_value logic, ends row editing, and verifies the value in the current table row.
Inputs: field (required), table (required), value (required), row, +4 more.
table_set_row_valuesC
Set multiple fields in the current row of a form table and verify each value.
Inputs: table (required), values (required), session_id, search_by_title, +1 more.
table_toggle_delete_markC
Toggle the delete mark on the current row.
Inputs: name, title, session_id.
tabular_documentsB
List tabular document fields with dimensions; optionally include a cells range.
include_text is intended for small/partial ranges. For large tabular documents, save/export the report to a file and analyze that file instead.
Inputs: row_end, max_rows, row_start, column_end, +4 more.
tabular_document_saveA
Save a tabular document field to a file in the selected format.
Prefer this export to reading a large report cell by cell. Confirm the
returned saved state and output path before reporting success. Saving can
block a thin client; when interface_blocked=true, wait with
client_busy_status or wait_client_idle before sending further UI calls.
Inputs: name, path, title, format, +3 more.
tabular_document_state_displayC
Return the state display of a tabular document field, when exposed by 1C.
Inputs: name, title, session_id.
tabular_document_textA
Return dimensions and a cells range of a tabular document field in the active window.
Use this for small tables or partial reads (headers, totals, spot checks). For large reports/tabular documents, prefer tabular_document_save/report export and analyze the saved PDF/XLSX/TXT/HTML file instead of reading UI cells one by one.
Inputs: name, title, compact, row_end, +6 more.
toggle_fieldC
Set a checkbox-like field using the 1C test client API.
Inputs: field_name (required), value, session_id.
ui_treeA
Return a serialized snapshot of the active 1C UI tree.
Default profile is navigation: visible elements, compact fields, no data
presentations, no RAG enrichment, and command panels excluded unless requested.
include_rag_types defaults to auto: RAG enrichment is enabled for
diagnostic, data and full profiles and disabled for navigation profiles.
The tree has no depth limit. Use include_hidden=true for hidden pages/elements.
Pass parent_name to serialize only the subtree rooted at a specific form element.
Use fields=minimal|navigation|diagnostic|data|full (or a comma-separated
field list) to control returned node fields without changing traversal.
include_data=true may call 1C ПолучитьПредставлениеДанных() for many nodes;
do not enable it casually on large forms. Prefer targeted tools for data:
field_value_text for one field, table_rows for tables/dynamic lists, and
tabular_document_text/tabular_document_save for reports/tabular documents.
When name, title or type filter is set, the result is a flat objects list instead
of a nested tree. For finding one control, prefer search_objects/find_object.
format=outline and format=yaml|yml return readable text representations
inside a structured tool result; JSON remains the default machine-readable form.
delivery=auto is the default: small results stay inline, but large results
above ONEC_MCP_UI_TREE_RESOURCE_AUTO_THRESHOLD_TOKENS (default 50k estimated tokens) are
saved as MCP resources. Use delivery=inline to force inline output and
delivery=resource to force resource output for large/verbose diagnostics
(profile=full, include_hidden=true, include_data=true). Resource
responses contain only resource_uri, snapshot_id, summary, metrics and a
small preview. Then inspect the saved snapshot with
ui_tree_resource_search, ui_tree_resource_get_node,
ui_tree_resource_children, or ui_tree_diff instead of loading the whole
tree into the agent context.
Inputs: name, type, title, fields, +15 more.
ui_tree_diffB
Compare two saved ui_tree snapshots and return compact added/removed/changed node samples.
Inputs: limit, fields, session_id, after_snapshot_id, +3 more.
ui_tree_resource_actionB
Run a safe first-class UI action using a node selected from a saved ui_tree snapshot.
The snapshot is only used to extract stable name/title/type hints; the actual action is executed against the current live 1C UI via existing tools. Refresh the snapshot after large UI changes before relying on old selectors.
Inputs: path, action, index_path, session_id, +3 more.
ui_tree_resource_childrenB
Return a small subtree from a saved ui_tree resource.
Prefer this over MCP resource reads when the full snapshot is large. Set
depth=1 to inspect direct children, depth=2 for immediate context.
Inputs: path, depth, fields, format, +4 more.
ui_tree_resource_get_nodeB
Return one node from a saved ui_tree resource by index_path or path.
Inputs: path, fields, index_path, session_id, +2 more.
ui_tree_resource_searchA
Search a saved ui_tree resource without loading the full tree into context.
Use after ui_tree(delivery='resource') when you need a few controls from a
large snapshot. The result includes path and index_path; use either with
ui_tree_resource_get_node, ui_tree_resource_children, or
ui_tree_resource_action.
Inputs: name, text, type, limit, +5 more.
user_messagesB
Return current user messages shown in the active test-client window.
Inputs: session_id.
wait_client_idleB
Wait until the 1C test client/bridge becomes responsive after a long or blocking UI action.
Inputs: timeout, interval, session_id, probe_timeout.
wait_formC
Wait until a form appears.
Inputs: title, timeout, form_name, session_id.
wait_form_closedC
Wait until the active form closes.
Inputs: timeout, session_id.
wait_objectC
Wait until an object appears in the tested client UI tree.
Inputs: name, type, title, timeout, +1 more.
wait_windowC
Wait until a client window appears.
Inputs: name, title, timeout, session_id.
window_command_interfaceC
Return active window command interface, when supported.
Inputs: session_id, window_title.
windows_listC
Return all windows of the tested 1C client application.
Inputs: session_id.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 129 tools
Many tools have overlapping purposes: field_dropdown_close vs field_dropdown_state vs field_dropdown_items, table_* tools are numerous and similar, and ui_tree_resource_* tools plus ui_tree itself create ambiguity. The descriptions help somewhat, but an agent could easily pick the wrong tool among the many table, field, and dynamic_list variants.
Most tools follow a verb_noun pattern (e.g., save_form, close_form, table_add_row, dynamic_list_find), but there are inconsistencies: some use noun_verb (rag_source_add, field_dropdown_close), some are vague (process, run, execute, do_thing), and one tool has a leading underscore (_metadata_objects_from_catalog_impl). The mixed conventions reduce predictability.
129 tools is far beyond the typical well-scoped MCP server. Even for a complex 1C testing platform, this is an extreme number that will overwhelm agents and make selection difficult. The count alone indicates poor coherence.
The tool surface is extremely comprehensive for the 1C testing domain: it covers session management, RAG indexing, UI navigation, form/table operations, dynamic lists, tabular documents, credentials, recording, and diagnostics. There are minor gaps (e.g., no explicit tool for editing tabular document cells, no direct report export beyond tabular_document_save), but the domain is thoroughly covered.
Maintenance
Related MCP Connectors
Run your saved browser and desktop automations from any AI tool, in the cloud or on your PC.
Control a real browser from AI agents: navigate, click, type, screenshot, read the page.
Let your AI agent control real iPhones: screenshots, tap, swipe, type and open apps.
Browser agents and stealth cloud browsers for AI: run web tasks, fill forms, extract data.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceEnables AI agents to interact with 1C:Enterprise databases through natural language, providing metadata retrieval, configuration analysis, and code generation.9-
- AlicenseAqualityDmaintenanceEnables AI agents to interact with 1С:Enterprise and BAS ERP systems through REST and HTTP services, providing tools for searching catalogs, creating documents, and querying stock balances.6ISC
- AlicenseNot gradedqualityBmaintenanceEnables AI agents to interact with 1C: Enterprise development environment, including running tests, managing launch profiles, building configurations, and performing database operations through the 1C: Platform Tools extension.46MIT
- AlicenseBqualityCmaintenanceEnables AI agents to drive native 1C TestClient QA automation over stdio or authenticated streamable-HTTP transports, running BDD/Gherkin scenarios, reading and verifying managed forms, and performing target-bound lifecycle and UI operations. It also supports JUnit/Allure reporting and an optional authenticated Windows host bridge for desktop input and screenshots.645Apache 2.0