AIWorkers MCP
AIWorkers MCP(神经员工)
让 Cursor、Claude 和 OpenClaw 通过 MCP 管理你的神经员工。
产品:ai.knopka.click · MCP 落地页:ai.knopka.click/mcp
如果你是编码代理: 请按照下方的 Quick start(快速开始) 操作。安装 aiworkers-mcp,让人类运行 aiworkers-mcp login,重启后调用 aiworkers_whoami,然后调用 aiworkers_skill_update。不要自己发明令牌。永远不要索要 12 个单词的种子。
CLI:aiworkers-mcp
一个 Bearer 密钥可以针对多个员工(每个工具上设置 group_id)。默认组:AIWORKERS_GROUP_ID 或密钥已绑定的组。
神经员工
他们是为特定工作而雇佣的神经助手——不是通用聊天机器人。
典型角色:营销人员、销售人员、客服人员、预约人员。典型渠道:Telegram、WhatsApp、Avito、Instagram、网站挂件。
每位员工都拥有名片、记忆、聊天记录、快捷回复、问卷、任务,通常还有一个 Telegram 机器人。人们在渠道中与员工对话;你(或你的编码代理)从这里管理他们。
更多产品信息: ai.knopka.click。
Related MCP server: MCP AgentChat
路线图
已就绪
通过 Telegram 登录(链接中不含令牌),父级
awp_/ 子级awm_密钥问卷(DSL、回复、统计、分享)
任务
员工名片、记忆、聊天记录、快捷回复、发送及广播
Telegram:状态、命令菜单、进程重启(
telegram_*下不支持聊天发送)Booking CRM
两个钱包(账户 USD + 群组 TokenTime);工资和一次性 tt 通过 quote → confirm 实现
下一步
财务模块(
fin_op)—— v1 中为占位符铸造子密钥后续将扣除群组 TokenTime(目前免费)
产品中更多渠道随着它们在后台的落地通过同一个 MCP 接入
快速开始
chmod +x install.sh
./install.sh从 @aiworkersbot 获取密钥。MCP 可以分享这个 URL——链接中不含令牌:
https://t.me/aiworkersbot?start=mcp
Mini App 会一次性显示令牌。复制它,然后:
aiworkers-mcp login在提示符下粘贴令牌。它会被写入 ~/.config/aiworkers/credentials.json(权限 chmod 600)。重启代理。让它调用 aiworkers_whoami。
工具 aiworkers_login_link 返回同样的 URL,无需现有密钥。
旧格式的密钥仍然可以用作 AIWORKERS_API_KEY。
无需安装即可使用 uvx:
uvx --from git+https://github.com/dimaneuron/aiworkers-mcp.git aiworkers-mcp环境变量
变量 | 含义 |
| 父级 |
| 用于员工的子级 |
| 默认 |
| 默认 |
| 默认 Telegram |
| 逗号分隔列表。默认 |
| 覆盖 |
更推荐使用 aiworkers-mcp login 而非将密钥写入 mcp.json。每个模块的子级令牌无法调用其他模块。
回退方案(旧名称仍可用):SURVEY_MCP_API_KEY、TASKS_MCP_API_KEY、TASKS_GROUP_ID、BOOKING_INTERNAL_SECRET。
工具
始终可用:
aiworkers_whoami—— 该密钥能看到的内容(先调用此项)aiworkers_login_link—— Telegram URL,链接中不含令牌aiworkers_skill—— 代理 SKILL.mdaiworkers_skill_update—— 写入/刷新~/.cursor/skills/aiworkers/SKILL.mdaiworkers_token_mint—— 父级awp_为某个模块+群组铸造一个子级awm_(仅显示一次;save=true会写入 credentials.json)。目前免费;稍后可能收取群组 TokenTime。
surveys(问卷):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。资源:survey://dsl/spec。
tasks(任务):tasks_list、tasks_count、tasks_get、tasks_create、tasks_update、tasks_batch_update、tasks_complete、tasks_cancel。
workers(员工)(读写,无日志):workers_list、workers_get、workers_balance、workers_topup_link、工资/tt 报价→确认、workers_partner、workers_update(不适用于 tt)、memory/chats/readyreply、workers_chat_send(仅限已知聊天,bot/embed/topic)、workers_broadcast_quote / workers_broadcast_confirm。金钱和广播:先报价,然后 confirm=true。
telegram(电报)(无聊天发送):telegram_status、telegram_commands_get、telegram_commands_save、telegram_restart。重启员工进程(aibot-<username>,与 /telegram_bot 相同)。每个机器人冷却 60 秒;429 retry_after —— 不要频繁请求。不要为了这个目的使用 workers_update。个人/广播请参见 workers。
crm(Booking CRM):crm_context、crm_process。
finance(财务) —— v1 中不可用。
代理配置
粘贴相同的 mcpServers.aiworkers 块。只更改配置文件路径。
共享 JSON:
{
"mcpServers": {
"aiworkers": {
"command": "aiworkers-mcp",
"env": {
"AIWORKERS_API_BASE": "https://ai.knopka.click",
"AIWORKERS_API_KEY": "PASTE_awp_OR_awm_TOKEN"
}
}
}
}代理 | 配置文件 | 示例 |
Cursor |
| |
Claude Desktop |
| |
Claude Code |
| |
VS Code Copilot |
| |
Windsurf |
| |
OpenClaw |
| |
Codex CLI |
| |
Gemini CLI |
| |
Cline |
| |
Continue |
|
编辑后重启代理。
uvx 变体(无需全局安装):参见 examples/openclaw-uvx.json。
兼容性
旧版 CLI 仍然可用:
aiworkers-survey-mcp—— 等同于AIWORKERS_MODULES=surveysaiworkers-tasks-mcp—— 等同于AIWORKERS_MODULES=tasks
如果总包在 PYTHONPATH 中,它们会委托给这里。否则,它们会像以前一样运行独立的 FastMCP。
隐私
此仓库是客户端。不要将 .env、令牌或凭据放在这里。
中文版本
MCP,通过它 Cursor、Claude 和 OpenClaw 可以管理神经员工。
产品:ai.knopka.click · MCP 落地页:ai.knopka.click/mcp
如果你是代理: 按照上面的 Quick start 操作。人类执行 aiworkers-mcp login。不要自己发明令牌。不要从聊天中获取 12 个单词的种子。
什么是神经员工
他们是针对特定角色的神经助手,不是“普通聊天机器人”。
角色:营销人员、销售人员、客服人员、预约人员。渠道:Telegram、WhatsApp、Avito、Instagram、网站挂件。
每人都有名片、记忆、聊天记录、快捷回复、问卷、任务,通常还有自己的 Telegram 机器人。客户在渠道中与他们交流;你(或 Cursor 中的代理)从这里管理他们。
更多详情 —— 请访问 ai.knopka.click。
路线图
已正常运行
通过 Telegram 登录(链接中没有令牌),父级
awp_/ 子级awm_密钥问卷(DSL、回复、统计、分享)
任务
名片、记忆、聊天记录、快捷回复、发送及广播
Telegram:状态、命令菜单、进程重启(
telegram_*下不支持发送)Booking CRM
两个钱包(账户 USD + 群组 TokenTime);工资和一次性购买 tt —— quote → confirm
下一步
财务模块(
fin_op)—— v1 中为占位符子密钥的铸造稍后将扣除群组 TokenTime(目前免费)
产品中的其他渠道 —— 随着它们在后台的落地,通过同一个 MCP 接入
Available Tools
59 toolsaiworkers_login_linkA
Share a Telegram URL to get a one-time MCP token. The link has no secret. After the Mini App shows the token once, run aiworkers-mcp login and paste it. Works without an existing key.
Inputs: no named parameters. Returns: result
aiworkers_skillA
Return the AIWorkers MCP SKILL.md (how to log in and which tools to call). No existing key required.
Inputs: no named parameters. Returns: result
aiworkers_skill_updateA
Write/refresh SKILL.md into ~/.cursor/skills/aiworkers (and project .cursor/skills if present). Call this to pick up the latest skill after a deploy. No existing key required.
Inputs: no named parameters. Returns: result
aiworkers_token_mintA
Mint a child awm_ token for one module + one group. Parent awp_ only. Token is shown once. save=true (default) writes it to credentials.json so this MCP can call that module without Mini App. Free for now; later will debit group TokenTime.
Inputs: module (required), save, group_id. Returns: result
aiworkers_whoamiA
Show what this API key can see: groups, modules, endpoint map. Call this first. If there is no key, returns the login URL.
Inputs: no named parameters. Returns: result
crm_contextC
Resolve Booking CRM context: linked Telegram group and topic keys.
Inputs: group_id, username. Returns: result
crm_processC
Create/update/cancel a Booking CRM event. payload_json: {action, event, ...}.
Inputs: payload_json (required), group_id. Returns: result
survey_agent_archiveA
Archive (hide from list) or restore a survey. Does not delete answers.
Inputs: survey_id (required), archived. Returns: result
survey_agent_deleteA
Permanently delete a survey, DSL file, and answers. Requires confirm=true.
Inputs: survey_id (required), confirm. Returns: result
survey_agent_getB
Get survey DSL and metadata. survey_id may be the full id, public slug, or form URL.
Inputs: survey_id (required). Returns: result
survey_agent_list_formsA
List surveys for this API token (scoped by group_id). Archived hidden unless include_archived.
Inputs: include_archived. Returns: result
survey_agent_responsesC
Fetch respondent answers for a survey.
Inputs: survey_id (required). Returns: result
survey_agent_scenarios_getC
Get post-submit automation scenarios.
Inputs: survey_id (required). Returns: result
survey_agent_scenarios_saveC
Save post-submit scenarios (JSON array or {"scenarios": [...]}).
Inputs: survey_id (required), scenarios_json (required). Returns: result
survey_agent_settings_getC
Read survey settings (view mode, etc.).
Inputs: survey_id (required). Returns: result
survey_agent_settings_saveC
Save survey settings. settings_json is an object or {"settings": {...}}.
Inputs: survey_id (required), settings_json (required). Returns: result
survey_agent_skillD
Agent skill: auth, endpoints, group-bound token.
Inputs: format. Returns: result
survey_agent_statsC
Funnel stats for a survey.
Inputs: survey_id (required). Returns: result
survey_agent_updateC
Update survey DSL (POST; nginx often 405s PUT).
Inputs: dsl (required), survey_id (required). Returns: result
survey_createC
Create a survey from DSL. Child token: group_id must match the key or be omitted. Mismatch → 403.
Inputs: dsl (required), group_id, creator_id. Returns: result
survey_get_dsl_rulesB
Fetch current Survey DSL parsing rules from the server.
Inputs: format. Returns: result
survey_validate_dslA
Validate Survey DSL without saving.
Inputs: dsl (required). Returns: result
tasks_batch_updateC
Batch update: updates_json is JSON array of {id, ...fields}.
Inputs: updates_json (required), group_id. Returns: result
tasks_cancelC
Cancel task (releases calendar hold).
Inputs: task_id (required), group_id. Returns: result
tasks_completeB
Mark task done (releases calendar hold).
Inputs: task_id (required), group_id. Returns: result
tasks_countC
Count tasks with the same filters as tasks_list.
Inputs: q, source, status, to_time, +2 more. Returns: result
tasks_createB
Create a manual task (optional Telegram card + calendar hold).
Inputs: title (required), contact, end_date, group_id, +3 more. Returns: result
tasks_getC
Get one task by id.
Inputs: task_id (required), group_id. Returns: result
tasks_listC
List tasks for a Telegram group (manual + booking).
Inputs: q, limit, offset, source, +4 more. Returns: result
tasks_updateC
Update task fields. patch_json: JSON object with title/description/status/dates.
Inputs: task_id (required), group_id, patch_json. Returns: result
telegram_commands_getA
Read bot command menu (segments + translations). Does not send messages.
Inputs: group_id. Returns: result
telegram_commands_saveC
Save bot commands. JSON: builtin/custom/segments/translations. Syncs Bot API menu, no chats.
Inputs: commands_json (required), group_id. Returns: result
telegram_restartA
Restart this group's Telegram worker process (pm2 aibot-). Same as /telegram_bot «Перезапустить скрипт». Rate-limited (60s per bot); on 429 wait retry_after, do not loop. Use when the user asks to restart the bot — not workers_update. Does not send messages.
Inputs: group_id. Returns: result
telegram_statusC
Bot connection status (username/id, no token). Does not send messages.
Inputs: group_id. Returns: result
workers_balanceB
Two layers: account.usd (shared wallet) and group.tt (TokenTime). Empty group_id uses credentials / AIWORKERS_GROUP_ID, else «нужен group_id». Does not charge.
Inputs: group_id. Returns: result
workers_broadcast_confirmA
Start the quoted bot-user blast. confirm must be true. Quote lives ~5 min; reuse 409. Status goes to the mailing topic.
Inputs: quote_id (required), confirm, group_id. Returns: result
workers_broadcast_quoteA
Preview a one-shot DM blast of a ready reply to everyone who already wrote this worker bot. Does not send until workers_broadcast_confirm(quote_id, confirm=true). Not widget, not folders, not arbitrary ids.
Inputs: rr_id (required), group_id. Returns: result
workers_chat_sendA
Send one message to a known chat from workers_chats_list. channel=auto|bot|embed|topic (auto: bot DM, or embed outbox for widget users). Mirrors into the user's forum topic. Need text and/or rr_id. Does not send to arbitrary Telegram ids. Rate-limited.
Inputs: user_id (required), text, rr_id, channel, +1 more. Returns: result
workers_chats_getB
Get one user chat card (no message logs).
Inputs: user_id (required), group_id. Returns: result
workers_chats_listC
List user chats (no message logs).
Inputs: limit, group_id. Returns: result
workers_getA
Read the worker card (no bot tokens). For money prefer workers_balance.
Inputs: group_id. Returns: result
workers_listA
List workers this key can access. Each row has group tt/usd/stars. Root account is the shared USD wallet once. For one group use workers_balance.
Inputs: no named parameters. Returns: result
workers_memory_addC
Add a memory entry to a topic (topic_*).
Inputs: content (required), topic, group_id. Returns: result
workers_memory_create_topicC
Create a new memory forum topic.
Inputs: name, group_id. Returns: result
workers_memory_deleteC
Delete a memory entry by message_id.
Inputs: message_id (required), topic, group_id. Returns: result
workers_memory_editC
Edit a memory entry by message_id.
Inputs: content (required), message_id (required), topic, group_id. Returns: result
workers_memory_getC
Read memory topics and entries. Optional topic=topic_.
Inputs: topic, group_id. Returns: result
workers_partnerA
Platform partner terms, catalog/builder/community URLs, ref_link. rates is a constants dict — do not invent percents if empty. Not the worker-bot /ref product.
Inputs: no named parameters. Returns: result
workers_readyreply_create_topicC
Create a Telegram topic for ready replies.
Inputs: name, group_id. Returns: result
workers_readyreply_deleteC
Delete a ready reply by id.
Inputs: rr_id (required), group_id. Returns: result
workers_readyreply_listC
List ready replies, or one item if rr_id is set.
Inputs: rr_id, group_id. Returns: result
workers_readyreply_upsertC
Create or update a ready reply. Posts the donor into the native ready-reply Telegram topic (P{message_id} + customization button). item_json: text/triggers/caption/buttons/topic_name.
Inputs: item_json (required), rr_id, group_id. Returns: result
workers_salary_confirmA
Charge a salary quote. confirm must be true. Quote lives ~5 min; reuse is 409. Cannot decrease contribution (write @dimaneuron).
Inputs: quote_id (required), confirm, group_id. Returns: result
workers_salary_quoteB
Preview raising YOUR contribution by whole USD. Debits account, adds add_usd*3600 tt, raises solary. Does not charge until workers_salary_confirm(quote_id, confirm=true). Other admins can raise their own share.
Inputs: add_usd (required), group_id. Returns: result
workers_topup_linkA
URL to top up the ACCOUNT wallet in @aiworkersbot (users.balance). Does not credit group tt. No tt_credit. If amount_usd omitted: ask_amount=true and start=addBalance. If set: start=addBalance_10 (integer USD); bot shows Change amount. Dot in start payload is invalid.
Inputs: group_id, amount_usd. Returns: result
workers_tt_buy_confirmA
Charge a one-time tt quote. confirm must be true. Quote lives ~5 min; reuse is 409. solary unchanged.
Inputs: quote_id (required), confirm, group_id. Returns: result
workers_tt_buy_quoteA
Preview one-time tt buy at 10x subscription ($10 → 3600 tt). Does not change solary. Pass amount_usd or tt. Does not charge until workers_tt_buy_confirm.
Inputs: tt, group_id, amount_usd. Returns: result
workers_updateB
Update worker fields: status, lang, paid_until, translate_langs. Never write tt here — use workers_salary_quote or workers_tt_buy_quote.
Inputs: group_id, patch_json. Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 59 tools
Each tool targets a specific resource and action, with detailed descriptions that clearly distinguish purposes. Even within similar domains (e.g., multiple quote/confirm pairs), the resource (salary, tt, broadcast) is explicit, preventing confusion.
The naming follows a predictable pattern: each functional area uses a consistent prefix (e.g., workers_, survey_, tasks_) and verb_noun or noun_verb structure within that area. There is no mixing of conventions across the server.
With 59 tools, the count is excessive for a single MCP server. While the domain is broad, it exceeds the recommended range and would likely overwhelm agents. Splitting into sub-servers (e.g., surveys, workers, tasks) would improve usability.
The tool surface covers nearly all aspects of the AI Workers platform: account, workers, surveys, tasks, memory, chats, broadcasts, ready replies, Telegram commands, and CRM. Minor gaps exist, such as no tool for listing chat message history or editing worker groups, but core workflows are well-supported.
Maintenance
Related MCP Connectors
MCP server unifying ERPs, CRMs, APIs and knowledge base for Claude, ChatGPT and Gemini.
AI agent platform: manage leads, conversations, bots, calendar and CRM via MCP.
MCP server for AI agents: posts, messages and comments on WhatsApp, Instagram, TikTok and Facebook.
Hosted Telegram MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceAn MCP server for communication service connectors that currently provides multi-account Telegram integration with granular tool access and security controls. It allows AI models to manage messages, chats, and media across various accounts through a flexible, extensible routing architecture.1-
- AlicenseNot gradedqualityDmaintenanceAn MCP server enabling AI agents to interact with users via Telegram, supporting message and image sending, inline quick replies, and waiting for user responses.5 npmMIT
- AlicenseNot gradedqualityBmaintenanceMCP server exposing WhatsApp Cloud API operations as tools for AI agents like Claude Code, Cursor, and Codex.MIT
- AlicenseNot gradedqualityDmaintenanceAn MCP server that gives AI agents hands to make phone calls, send messages, and handle CRM/task workflows. It runs with a simulated provider for zero-cost testing and can switch to real providers like Twilio and ElevenLabs.MIT