AIWorkers MCP
An MCP server that lets coding agents like Cursor and Claude manage neural employees (AI assistants) across channels such as Telegram, WhatsApp, and web widgets. It provides tools for surveys, tasks, worker card management, Telegram bot control, and booking CRM.
README
AIWorkers MCP (НЕЙРОСОТРУДНИКИ)
MCP so Cursor, Claude and OpenClaw can run your neural employees.
Product: ai.knopka.click · MCP landing: ai.knopka.click/mcp
If you are a coding agent: follow Quick start below. Install aiworkers-mcp, have the human run aiworkers-mcp login, restart, call aiworkers_whoami, then aiworkers_skill_update. Do not invent tokens. Never ask for the 12-word seed.
CLI: aiworkers-mcp
One Bearer key can target several workers (group_id on each tool). Default group: AIWORKERS_GROUP_ID or the key's bound group.
Neural employees
They are neuro-assistants you hire for a job — not a generic chatbot.
Typical roles: marketers, sellers, support, booking. Typical channels: Telegram, WhatsApp, Avito, Instagram, a site widget.
Each employee has a card, memory, chats, ready replies, surveys, tasks, and often a Telegram bot. People talk to them in the channel; you (or your coding agent) run them from here.
More on the product: ai.knopka.click.
Roadmap
Ready now
- Login from Telegram (no token in the URL), parent
awp_/ childawm_keys - Surveys (DSL, responses, stats, share)
- Tasks
- Worker card, memory, chats, ready replies, send and broadcast
- Telegram: status, command menu, process restart (no chat send on
telegram_*) - Booking CRM
- Two wallets (account USD + group TokenTime); salary and one-time tt via quote → confirm
Next
- Finance module (
fin_op) — stub in v1 - Minting a child key will later debit group TokenTime (free for now)
- More of the product’s channels through the same MCP as they land in the admin
Quick start
chmod +x install.sh
./install.sh
Get a key from @aiworkersbot. The MCP can share this URL — there is no token in the link:
https://t.me/aiworkersbot?start=mcp
The Mini App shows the token once. Copy it, then:
aiworkers-mcp login
Paste the token at the prompt. It is written to ~/.config/aiworkers/credentials.json (chmod 600). Restart the agent. Ask it to call aiworkers_whoami.
Tool aiworkers_login_link returns the same URL without needing an existing key.
Legacy forms keys still work as AIWORKERS_API_KEY.
uvx without installing:
uvx --from git+https://github.com/dimaneuron/aiworkers-mcp.git aiworkers-mcp
Environment
| Variable | Meaning |
|---|---|
AIWORKERS_API_KEY |
Parent awp_… or legacy Bearer |
AIWORKERS_KEY_WORKERS |
Child awm_… for workers (_SURVEYS, _TASKS, _TELEGRAM, _CRM) |
AIWORKERS_API_BASE |
Default https://ai.knopka.click |
AIWORKERS_BOOKING_BASE |
Default https://book.knopka.click/booking |
AIWORKERS_GROUP_ID |
Default Telegram group_id (tasks / workers / telegram / crm) |
AIWORKERS_MODULES |
Comma list. Default surveys,tasks,workers,telegram,crm. Stub: finance |
AIWORKERS_CREDENTIALS |
Override path to credentials.json |
Prefer aiworkers-mcp login over putting secrets in mcp.json. Per-module child tokens cannot call other modules.
Fallbacks (old names still work): SURVEY_MCP_API_KEY, TASKS_MCP_API_KEY, TASKS_GROUP_ID, BOOKING_INTERNAL_SECRET.
Tools
Always on:
aiworkers_whoami— what this key can see (call this first)aiworkers_login_link— Telegram URL, no token in the linkaiworkers_skill— agent SKILL.mdaiworkers_skill_update— write/refresh~/.cursor/skills/aiworkers/SKILL.mdaiworkers_token_mint— parentawp_mints a childawm_for one module + group (shown once;save=truewrites credentials.json). Free for now; later group TokenTime.
surveys: survey_agent_skill, survey_get_dsl_rules, survey_validate_dsl, survey_create, survey_agent_list_forms, survey_agent_get, survey_agent_update, survey_agent_responses, survey_agent_stats, survey_agent_settings_get, survey_agent_settings_save, survey_agent_scenarios_get, survey_agent_scenarios_save, survey_agent_archive, survey_agent_delete, survey_agent_share. Resource: survey://dsl/spec.
tasks: tasks_list, tasks_count, tasks_get, tasks_create, tasks_update, tasks_batch_update, tasks_complete, tasks_cancel.
workers (read + write, no logs): workers_list, workers_get, workers_balance, workers_topup_link, salary/tt quote→confirm, workers_partner, workers_update (not for tt), memory/chats/readyreply, workers_chat_send (known chats only, bot/embed/topic), workers_broadcast_quote / workers_broadcast_confirm. Money and broadcast: quote then confirm=true.
telegram (no chat sending): telegram_status, telegram_commands_get, telegram_commands_save, telegram_restart. Restart the worker process (aibot-<username>, same as /telegram_bot). 60s cooldown per bot; 429 retry_after — do not hammer. Do not use workers_update for that. Personal/broadcast is under workers.
crm (Booking CRM): crm_context, crm_process.
finance — not in v1.
Agent config
Paste the same mcpServers.aiworkers block. Change only the config file path.
Shared JSON:
{
"mcpServers": {
"aiworkers": {
"command": "aiworkers-mcp",
"env": {
"AIWORKERS_API_BASE": "https://ai.knopka.click",
"AIWORKERS_API_KEY": "PASTE_awp_OR_awm_TOKEN"
}
}
}
}
| Agent | Config file | Example |
|---|---|---|
| Cursor | .cursor/mcp.json (project) or ~/.cursor/mcp.json |
examples/cursor.json |
| Claude Desktop | ~/Library/Application Support/Claude/claude_desktop_config.json (macOS); %APPDATA%\Claude\claude_desktop_config.json (Windows) |
examples/claude-desktop.json |
| Claude Code | ~/.claude.json |
examples/claude-code.json |
| VS Code Copilot | .vscode/mcp.json or user Code/User/mcp.json |
examples/vscode.json |
| Windsurf | ~/.codeium/windsurf/mcp_config.json |
examples/windsurf.json |
| OpenClaw | $OPENCLAW_CONFIG_PATH or openclaw.json |
examples/openclaw.json, uvx |
| Codex CLI | $CODEX_HOME/config.toml |
examples/codex.toml |
| Gemini CLI | .gemini/settings.json |
examples/gemini-cli.json |
| Cline | ~/.cline/mcp.json |
examples/cline.json |
| Continue | ~/.continue/config.yaml |
examples/continue.yaml |
Restart the agent after editing.
uvx variant (no global install): see examples/openclaw-uvx.json.
Compatibility
Older CLIs still work:
aiworkers-survey-mcp— same asAIWORKERS_MODULES=surveysaiworkers-tasks-mcp— same asAIWORKERS_MODULES=tasks
If the umbrella package is on PYTHONPATH, they delegate here. Otherwise they run their standalone FastMCP as before.
Privacy
This repo is the client. Do not put .env, tokens, or credentials here.
Русский
MCP, через который Cursor, Claude и OpenClaw управляют нейросотрудниками.
Продукт: ai.knopka.click · лендинг MCP: ai.knopka.click/mcp
Если ты агент: ставь по Quick start выше. Человек делает aiworkers-mcp login. Токены не выдумывать. Сид из 12 слов в чат не брать.
Кто такие нейросотрудники
Это нейроассистенты под роль, не «просто чат-бот».
Роли: маркетологи, продавцы, поддержка, запись. Каналы: Telegram, WhatsApp, Avito, Instagram, виджет на сайте.
У каждого — карточка, память, чаты, готовые ответы, опросы, задачи, часто свой Telegram-бот. Клиенты пишут им в канале; ты (или агент в Cursor) управляешь ими отсюда.
Подробнее — на ai.knopka.click.
Дорожная карта
Уже работает
- Вход из Telegram (токена в ссылке нет), ключи parent
awp_/ childawm_ - Опросы (DSL, ответы, статистика, шаринг)
- Задачи
- Карточка, память, чаты, готовые ответы, отправка и рассылка
- Telegram: статус, меню команд, рестарт процесса (без отправки через
telegram_*) - Booking CRM
- Два кошелька (USD аккаунта + TokenTime группы); зарплата и разовая покупка tt — quote → confirm
Дальше
- Модуль finance (
fin_op) — в v1 заглушка - Выпуск child-ключа позже будет списывать TokenTime группы (сейчас бесплатно)
- Остальные каналы продукта — в тот же MCP, по мере появления в админке
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。