Yodu AI Employees MCP

Name: Yodu AI Employees MCP
Author: noveum-ai

Connect your assistant to the AI team doing your company’s work.

От сообществаРаботаетБез входаГлобальныйБесплатноМожет изменять данные

Что умеет

Get Workspace: Returns the workspace (organization) this connection is bound to.
Get Workspace Overview: High-level activity for the workspace over the last N days (agents, tasks, needs-attention).
Get Workspace Settings: Returns safe workspace-level model routing and runtime resource limits. Owner/admin only; no credentials or raw runtime payloads are returned.

Какие данные видит

Нужен ли аккаунт

Не нужен: сервер работает без входа

Connect your assistant to the AI team doing your company’s work. Yodu exposes workspace-scoped tools to inspect employees and tasks, create and update work, manage memory, and configure supported schedules and settings.

Use it to check what needs attention, assign a research task, or inspect an employee’s progress from your MCP client. Begin with get_workspace to confirm the connected workspace. User roles and tool permissions apply.

This listing uses Yodu’s OAuth endpoint: https://app.yodu.ai/api/mcp/oauth. API-key clients can use https://app.yodu.ai/api/mcp with a workspace-bound bearer key. Setup: https://docs.yodu.ai/developers/client-setup.

Yodu is a hosted commercial service: $499/month includes personal setup and up to five active AI employees, including a Chief of Staff. There is no free trial. Model-provider and third-party tool costs are separate. The MCP connection operates Yodu; employee model access is configured separately.

Как подключить

Как подключить Yodu AI Employees MCP к Claude

Адрес сервера, который нужно вставить в настройках Claude:

https://yodu--noveum-ai.run.tools

Откройте Claude (сайт claude.ai или приложение для компьютера).
Перейдите в Настройки → Коннекторы (в английском интерфейсе: Settings → Connectors).
Нажмите Добавить пользовательский коннектор (Add custom connector).
Вставьте адрес сервера, скопированный ниже, в поле URL удалённого MCP-сервера и нажмите Добавить.
В чате нажмите + → Коннекторы и включите новый коннектор.

Пользовательские коннекторы доступны на тарифах Free, Pro, Max, Team и Enterprise. На Free можно добавить только один. В Team и Enterprise коннектор сначала добавляет владелец организации в Настройки организации → Коннекторы.

Авторизация

После нажатия Добавить откроется окно входа в сервис. Войдите под своим аккаунтом и подтвердите доступ. Пароль Claude не видит.

Как подключить Yodu AI Employees MCP к ChatGPT

Адрес сервера, который нужно вставить в настройках ChatGPT:

https://yodu--noveum-ai.run.tools

Откройте ChatGPT в браузере (chatgpt.com). Нужен тариф Plus, Pro, Business, Enterprise или Edu.
Один раз включите режим разработчика: Настройки → Приложения → Дополнительные настройки → Режим разработчика (Settings → Apps → Advanced settings → Developer mode).
Откройте Настройки → Коннекторы и нажмите Создать (Create).
Заполните форму: Название (любое), Описание (одна фраза, что делает сервис), URL MCP-сервера (скопируйте ниже).
В поле Аутентификация выберите OAuth, если сервис требует вход, иначе Нет. Нажмите Создать.
В новом чате откройте + → Приложения/Коннекторы и включите коннектор.

Название раздела у OpenAI менялось (Connectors → Apps/Plugins). Если пункт называется иначе, ищите «режим разработчика» в настройках. В рабочих пространствах Business/Enterprise администратор должен разрешить пользовательские коннекторы.

Авторизация

При первом использовании ChatGPT откроет окно входа в сервис. Войдите и подтвердите доступ.

Как подключить Yodu AI Employees MCP к Cursor

Адрес сервера, который нужно вставить в настройках Cursor:

https://yodu--noveum-ai.run.tools

Быстрый способ: нажмите кнопку Открыть в Cursor ниже, Cursor сам предложит добавить сервер.

Вручную:

В Cursor откройте Settings → Cursor Settings → MCP и нажмите Add new global MCP server.
Вставьте JSON, скопированный ниже, в открывшийся файл ~/.cursor/mcp.json (для одного проекта: .cursor/mcp.json в корне проекта).
Сохраните файл. Сервер появится в списке MCP; при необходимости пройдите авторизацию по кнопке рядом с ним.

Авторизация

Если сервис требует вход, рядом с сервером в списке MCP появится кнопка авторизации.

Список инструментов сервера (60)

Технические названия из tools/list. Нужны только разработчикам.

get_workspace	Returns the workspace (organization) this connection is bound to.
get_workspace_overview	High-level activity for the workspace over the last N days (agents, tasks, needs-attention).
get_workspace_settings	Returns safe workspace-level model routing and runtime resource limits. Owner/admin only; no credentials or raw runtime payloads are returned.
list_agents	Lists every agent in the workspace (any status) with its runtime id and channel.
list_secrets	Lists the workspace's configured secrets by NAME and scope, never their values. Each entry is { scope: 'workspace' \| 'agent', agentId?, provider, name, updatedAt, reachesAgentSandbox }. A 'workspace' secret is shared by every agent; an 'agent' secret is private to that agent and overrides the workspace secret of the same name. `reachesAgentSandbox` is the answer to 'can the agents themselves read this?': false means the platform uses it on their behalf but it is absent from their environment, so a script doing `$NAME` will find nothing. Set it with `exposeToAgents` on set_workspace_secret. It reports the CONFIGURED state: sandbox environment is fixed when a container is created, so an agent that is mid-task keeps the old environment until its sandbox is next recreated. Use this to see what is already configured before setting or rotating a secret. Secret VALUES can never be read back through this API.
list_role_templates	Lists the role templates a new agent can be created from (id, name, description).
list_agent_configs	Lists versioned employee configuration documents, including their Markdown bodies. Restricted to workspace owners and admins because these are privileged operating instructions.
list_models	Lists the model references currently runnable in this workspace. Use a returned ref with set_agent_model. Owner/admin only because provider configuration is workspace settings metadata.
get_agent_skills	Returns one employee's role-default skills, explicit add/remove override, resolved loadout, assignable catalog, and protected baseline. agentId accepts either the agent id from list_agents or its runtimeAgentId handle (for example program-manager). Owner/admin only.
list_connected_tools	Lists safe connected-app metadata plus one employee's current tool-grant rows. agentId accepts either the agent id from list_agents or its runtimeAgentId handle (for example program-manager). Secret values and internal connection config are never returned. Owner/admin only.
list_agent_files	Lists metadata for versioned custom employee files under safe skills/ and scripts/ paths. File content is never returned. Owner/admin only.
list_memory	Lists workspace memory with optional lifecycle/scope filters. Defaults to shared ACTIVE and DRAFT documents. Employee-local memory must be requested separately with exactly AGENT_LOCAL_MEMORY plus agentId and is owner/admin only.
read_memory	Reads one memory document by id. Shared memory is member-readable; set includeAgentLocal only for an employee-local ID (owner/admin only).
activate_memory	Activates a sourced non-governed memory draft as live employee context after validating that it has content and provenance. Company OS governed mirrors are read-only and change through proposal review. Owner/admin only; queues runtime application.
set_memory_status	Moves a non-governed memory document to DRAFT, SUPERSEDED, or ARCHIVED without deleting it. Company OS governed mirrors are read-only and change through proposal review. Use activate_memory for ACTIVE so provenance validation cannot be bypassed. Owner/admin only.
list_tasks	Lists board tasks, optionally filtered by agent, status, priority, label, or search.
read_task	Reads one workspace task with its discussion, approval summaries, evidence-file metadata, and activity timeline. Approval decisions and file contents remain outside this tool.
get_runtime_health	Latest runtime health snapshot and last successful file-sync time for the workspace.
get_runtime_state	Current workspace runtime connection, container, queue, backup, and sync state. Raw commands, errors, endpoints, host paths, worker identities, and credentials are never returned.
get_runtime_command_status	Reads the safe status and retry timing for one workspace runtime command. Payloads, results, raw errors, worker identities, host details, and credentials are never returned.
list_custom_mcp_servers	Lists the workspace's own custom MCP servers: id, name (the mcporter key), URL, description, configured header NAMES, and the tools last enumerated from each server. Secret header values are never returned. Use the id with remove_custom_mcp_server and the name with set_agent_tool_access.
list_scheduled_jobs	Lists one cursor-paginated page of the workspace's scheduled agent jobs with status, cron expression, next/last run, assigned agent, and recent run history. Job prompts are included in the response, so any key with workspace:read can view them.
create_memory	Creates a memory document. Shared scopes are member-writable; AGENT_LOCAL_MEMORY requires agentId and an owner/admin role. Body is Markdown.
update_memory	Updates a non-governed memory document's title/summary/body by id. Company OS governed mirrors are read-only and change through proposal review. ACTIVE edits must still satisfy provenance/content invariants and are immediately re-applied to the runtime; for reviewed replacements, create a draft and call activate_memory. Set includeAgentLocal only for an employee-local ID (owner/admin only).
create_task	Creates a task on the workspace board.
update_task	Updates a workspace task's content, priority, due date, link, assignee, or status. Assignment and material changes wake the assigned employee through the existing task workflow.
move_task	Moves a task to a board column and optionally positions it between two same-column tasks. A real status transition wakes the assignee; same-column reordering stays silent.
comment_task	Adds an authenticated user comment or threaded reply to a workspace task, then wakes its assigned employee (or routes an unassigned task through the existing workflow).
create_agent	Creates a safe recruitment draft from a role template. Use the dedicated model, skill, tool-access, and versioned-config tools after creation; raw internal config is intentionally not accepted or returned. Use provision_agent to bring it online.
provision_agent	Provisions an agent onto the runtime so it can receive and reply to messages.
repair_workspace_runtime	Force-rebuilds this workspace's runtime from Yodu's saved config: a safety backup is taken first, then the container stack is recreated and the full bundle (openclaw.json plus every agent's files) is regenerated from the database. Use when the runtime has drifted or broken, or to roll out config changes to every agent at once. Disruptive: in-flight agent turns are interrupted while the stack restarts.
set_workspace_model_routing	Sets the workspace primary model and ordered cross-provider fallbacks, then queues runtime application. Every configured provider must appear exactly once in a custom route; null primary with no fallbacks restores automatic routing.
set_workspace_runtime_resources	Sets workspace gateway memory/CPU limits and queues runtime application. Empty/null restores platform defaults. Owner/admin only.
set_workspace_cost_guardrail	Sets the work-admission cost ceiling in USD for the rolling 3-HOUR admission window; agents stop creating NEW tasks once consumption in that window crosses it (work already in flight and its review are never suppressed). Measures list-price consumption INCLUDING flat-subscription usage, because a subscription-only workspace has no invoice to guard — so set it against 3 hours of list-price burn, which on Opus-class models routinely runs to three figures, not against a monthly invoice. Pass null to remove the ceiling. Owner/admin only.
set_agent_autonomy	Sets an agent's autonomy mode (DRAFT_ONLY, APPROVAL_REQUIRED, AUTONOMOUS). AUTONOMOUS_UNSAFE (skips ALL human approval gates) is intentionally not settable via API. Use the app UI.
write_agent_config	Writes one of the seven versioned employee configuration documents: SOUL, SYSTEM_PROMPT, MEMORY_POLICY, TOOL_POLICY, RUNTIME_FILES, SCHEDULE, or MODEL_POLICY. `body` is Markdown. Set status ACTIVE to apply it to the running employee; DRAFT to stage it.
set_agent_model	Sets the model an agent runs on (e.g. claude-max/claude-opus-4-8, openai/gpt-5.5). Must be a model configured for this workspace. Pass null/empty to clear the override and use the workspace default.
set_agent_role	Reassigns an agent to a different role. Pass the target role's roleId (e.g. "customer-support", from list_role_templates); the role must already exist in this workspace. The stored role is normalized to that roleId, so this also fixes an agent whose role was set to a raw template id. Identity handles (channelSlug, runtimeAgentId) are not changed.
set_agent_skills	Adjusts an agent's skill loadout: `add` and/or `remove` skill slugs relative to its role's baseline.
set_agent_cloud_coding	Enables or disables Claude Code or Codex for one employee. Disabling uses the normal cancellation-safe workflow and stops that employee's live sessions for the provider.
set_agent_tool_access	Enables or disables ONE tool integration for ONE agent. Tool access is a per-agent boolean: enabled (the default when no grant exists) means the agent sees and can use the tool; disabled hides it from that agent entirely. For a connected app pass connectionId plus category = the provider and integrationKey = composio:<provider>; toolName defaults to the connection's display name. For a workspace custom MCP server pass category = custom-mcp, integrationKey = custom-mcp:<serverId>, toolName = the server name from list_custom_mcp_servers (required: the runtime hides MCP servers by name; the id is also accepted); never pass connectionId with custom-mcp. Risky actions on enabled tools still go through the approval gate, which is controlled per agent by set_agent_autonomy, not here.
create_scheduled_job	Creates a recurring job: at each cron tick the prompt is delivered to the assigned agent's channel as a message to act on. cronExpression is standard 5-field cron evaluated in the job's timezone (an IANA name such as America/New_York; missing or invalid values fall back to UTC), so 0 15 * * 1-5 with timezone America/New_York fires at 15:00 New York time on weekdays. Pass agentId to assign the job; an unassigned job is skipped at run time.
update_scheduled_job	Updates a scheduled job. Pass the job's `id`, the value returned by create_scheduled_job or list_scheduled_jobs (`jobId` is accepted as an alias). Updates name, description, cronExpression, timezone, prompt, agentId, or status (PAUSED, ARCHIVED). Activation is deliberately separate and parity-gated. A new cronExpression or timezone recomputes timing in the job's effective timezone (an IANA name such as America/New_York; missing or invalid values fall back to UTC).
delete_scheduled_job	Deletes a scheduled job by id (from list_scheduled_jobs). Returns deleted: false when no such job exists in this workspace.
run_scheduled_job	Runs one scheduled job immediately for end-to-end testing without changing its cron cadence. This is model-initiated work, so the global autonomous-execution gate and the employee's normal approval policy both apply.
activate_scheduled_job	Activates a PAUSED or draft scheduled job so its cron fires again. Refused when the assigned employee is absent from the latest runtime parity check, because a schedule pointing at an agent the runtime does not have would fire into nothing. Pair with update_scheduled_job, which can pause but deliberately cannot activate.
add_custom_mcp_server	Registers a workspace-owned custom MCP server (name + https URL + optional request headers) that every agent can reach via mcporter. Header values are stored encrypted and never returned.
remove_custom_mcp_server	Removes a workspace-owned custom MCP server. Pass the server's `id` from list_custom_mcp_servers (`serverId` is accepted as an alias). Queues a runtime provision so it disappears from every agent's mcporter registry. The stored list, minus that server and its encrypted request headers, is saved as a new secret version. Removing an unknown id is a no-op.
set_model_provider_key	Stores a model provider API key (OpenAI/Anthropic/OpenRouter/Claude subscription). Write-only: it can never be read back through MCP.
connect_cloud_agent	Connects a Claude Code routine (created at claude.ai/code/routines) so AI employees can delegate coding tasks to cloud sessions on the customer's Claude subscription. Write-only: the trigger token can never be read back.
list_connected_github_for_coding	Lists workspace-scoped GitHub integrations and client-safe repository metadata for cloud coding. Connected-account identifiers, credentials, and tokens are never returned. Owner/admin only.
connect_cloud_coding_repo	Connects a GitHub repo for INTERACTIVE cloud coding (the primary tier): a real Claude Code session runs in a per-workspace container, clones the repo, works on a branch, and opens a PR. Unlike connect_cloud_agent (routines), this needs NO routine id and NO token. For a private repo, first call list_connected_github_for_coding and pass its workspace-scoped integration connection id; the Yodu GitHub App must already have an active owner-verified binding to this workspace.
list_cloud_agent_connections	Lists connected cloud coding agents — both routines and interactive sidecar repos. Tokens are never included.
update_cloud_agent_connection	Revokes one cloud-agent or cloud-coding connection in this workspace while preserving its audit history. The organization fence is part of the update, and no secret or credential is returned. Owner/admin only.
get_cloud_agent_setup_guide	Returns Yodu's canonical Claude Code routine prompt template and the numbered claude.ai steps to set up a cloud coding agent (plan requirements, GitHub access, routine creation, Environment and network configuration, and the API trigger). Read-only: use connect_cloud_agent once you have the routine's trig_ id and trigger token. There is no routine-creation API, so a human must perform the claude.ai steps; this tool only returns the instructions.
set_workspace_secret	Stores an arbitrary tool secret (e.g. APOLLO_API_KEY, a Stripe restricted key). Write-only: it can never be read back through MCP. By DEFAULT a secret is consumed by the platform on the agent's behalf (model providers, web search, MCP auth) and is NOT visible to the agent itself. Pass `exposeToAgents: true` to also project the value into every agent's sandbox environment, so their own scripts and shell commands can read it as `$NAME` (the database URLs the `db-readonly-lookups` skill reads, YODU_AGENT_DB_URL_RO / YODU_AGENT_DB_URL / API_MARKET_DB_URL, are projected without the flag). The agent runtime DELETES sandbox variables whose names look like a provider credential, so `exposeToAgents` is refused for a name ending in API_KEY, TOKEN, PASSWORD, PRIVATE_KEY or SECRET: put such a credential in a single connection URL instead (ES_URL rather than ES_HOST + ES_PASSWORD). Call `list_secrets` afterwards to confirm `reachesAgentSandbox`. A newly exposed value reaches an agent when its sandbox container is next created, not inside a turn that is already running. Pass `agentId` to scope the secret to ONE agent (it overrides the workspace secret of the same name for that agent); omit it for a workspace-wide secret shared by every agent. Use an env-style NAME (e.g. TWITTER_API_KEY) so it becomes a valid environment variable.
write_agent_file	Provisions a custom file into ONE agent's runtime workspace: a helper script the agent runs (e.g. scripts/x_post.py) or a custom skill (skills/<name>/SKILL.md). `path` is relative and MUST start with `skills/` or `scripts/`; no root-level file is writable (`.env`, SOUL.md, MEMORY.md, TOOLS.md, IDENTITY.md, USER.md, HEARTBEAT.md, AGENTS.md are owned by the runtime/config docs) and the path can never escape the workspace. The baseline skill directories (skills/mcporter/, skills/human-writing/, skills/db-readonly-lookups/) are platform-managed and not writable either. The file is written on the next provision. Returns metadata only.
list_management_grants	Lists who currently holds which Company OS management authority (WORK_TRANSITION, WORK_ASSIGN, REVIEW_DECIDE, ...) in this workspace, with the agent or member each grant belongs to. Revoked and expired grants are omitted. Read this before granting: an authority somebody already holds does not need a second grant.
update_management_grant	Issues or withdraws one Company OS management authority, mirroring the workspace UI. operation="GRANT" needs principalId + action; operation="REVOKE" needs grantId + reason. Grant WORK_TRANSITION to let a teammate accept a peer's submitted work, which is what drains a REVIEW queue that has paused task creation workspace-wide. `principalId` accepts an agent id or its runtimeAgentId handle. Scope defaults to the whole workspace; only AGENT scope takes a boundary, and it requires boundaryAgentId. A withdrawn grant is not resurrected by the role reconcile; issue a new one to restore the authority. Revoking WORK_CREATE or WORK_ASSIGN does not stop an agent handing work to a hired teammate, which every agent may do, or handing over a task it owns. Owner/admin only.
list_workspaces	Lists the workspaces this connection is authorized to reach and marks which one is the active workspace. Every other tool acts on the active workspace only. Call set_active_workspace to move to a different one.