This server lets you fully manage AI workers, tasks, knowledge, integrations, and account resources on the CreateWorker platform.
Account & Usage
Retrieve organization details and credit wallet balance
Browse paginated credit usage ledger entries
Worker Management
List, get, create, update (name/status), and soft-archive AI workers
Task Management
List, get, create, cancel, and wait for completion of tasks assigned to workers
Retrieve all deliverables (outputs) produced by a task, or fetch a specific deliverable by ID
Approvals & Clarifications
Approve, reject, or edit proposed worker actions (human-in-the-loop)
Answer clarifying questions from workers by selecting an option
Knowledge Base
List, get, create, and delete knowledge documents to serve as reference material for workers
Integrations & Connections
List, create, and delete connections to external APIs (outbound) or inbound webhook endpoints
Bind connections to workers (inbound, outbound, or both) and manage those bindings
Event Webhooks
Register URLs to receive notifications for CreateWorker platform events (e.g. task.completed, proposal.needs_approval)
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., "@@createworker/mcp Create a task for my content worker to write a blog 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.
MCP (Model Context Protocol) server for CreateWorker. Drive AI workers — create and run tasks, read deliverables, approve actions, manage knowledge, and connect your own apps — directly from Claude Code, Claude Desktop, Cursor, and any other MCP client.
Requires Node 18+ and a CreateWorker API key (create one in your dashboard under Developer).
Local (npx):
claude mcp add createworker --env CREATEWORKER_API_KEY=cw_live_… -- npx -y @createworker/mcpRemote (hosted, zero-install):
claude mcp add --transport http createworker https://www.createworker.com/api/mcp \
--header "Authorization: Bearer cw_live_…"Related MCP server: ClickUp MCP
Add to claude_desktop_config.json:
{
"mcpServers": {
"createworker": {
"command": "npx",
"args": ["-y", "@createworker/mcp"],
"env": { "CREATEWORKER_API_KEY": "cw_live_…" }
}
}
}Add to .cursor/mcp.json (same command/args/env shape as above).
Env var | Required | Default |
| yes | — |
| no |
|
Workers (workers_list, worker_get/create/update/delete — create/update take externalExecutor) · Tasks (tasks_list, task_get/create/cancel/wait/deliverables) · External-worker runner (task_claim, task_progress, deliverable_submit) · Chat (chat_pending, chat_history, chat_reply) · approval_decide · clarification_answer · deliverable_get · Knowledge (knowledge_list/get/create/delete) · Integrations — connect your own app (connections_list, connection_create_api, connection_create_webhook, connection_delete, bindings_list, binding_create, binding_delete) · webhook_register · account_get, usage_list.
A key's scopes and each worker's autonomy level still bound what actually executes — write actions stay approval-gated unless a worker is set to auto-execute.
"List my CreateWorker workers, then create a task for the support worker to summarize today's tickets and wait for the result."
"Connect my site's API (https://api.example.com, bearer token) and bind it to my content worker so it can publish posts."
"Be the runner for my external worker: poll tasks_list for ASSIGNED tasks, task_claim the oldest, do the work, and deliverable_submit the result. Also answer any chat_pending messages with chat_reply." (see the External Workers guide)
MIT
Get the API key's organization and credit wallet balance.
Inputs: no named parameters.
Decide a PROPOSED task's action: APPROVE (execute), REJECT, or EDIT (provide editedAction).
Inputs: taskId (required), decision (required), editedAction.
Bind a connection to a worker so the worker can use it. purpose: OUTBOUND (worker calls your API), INBOUND (your app triggers the worker), or BOTH.
Inputs: workerId (required), connectionId (required), purpose.
Remove a worker↔connection binding by id.
Inputs: bindingId (required).
List worker↔connection bindings, optionally for one worker.
Inputs: workerId.
Answer a worker's clarifying question (CLARIFICATION_PENDING) by selecting an option.
Inputs: taskId (required), proposalId (required), selectedOptionKey (required), note.
Create a generic_api connection so a bound worker can call YOUR app's API (outbound). allowedHosts is the required SSRF allowlist (e.g. ['api.example.com']).
Inputs: displayName (required), allowedHosts (required), apiKey, baseUrl, +5 more.
Create a webhook connection: returns an endpointUrl + signingSecret (once) your app POSTs events to, which become tasks for the bound worker.
Inputs: displayName (required), generateSecret, inboundAuthMode.
Disconnect an integration connection by id.
Inputs: connectionId (required).
List integration connections (generic API + webhook).
Inputs: limit, cursor.
Get a single deliverable by id, including its full content.
Inputs: deliverableId (required).
Create a text knowledge document (chunked/embedded asynchronously). Use this to give workers reference material — e.g. your API docs.
Inputs: title (required), content (required), categoryId.
Delete a knowledge document by id.
Inputs: docId (required).
Get a knowledge document by id.
Inputs: docId (required).
List the org's knowledge documents (paginated).
Inputs: limit, cursor.
Cancel a task that has not yet executed.
Inputs: taskId (required).
Create a task for a worker. Runs asynchronously — then poll with task_get or use task_wait. Writes stay approval-gated unless the worker auto-executes.
Inputs: title (required), workerId (required), payload, description.
List the deliverables a task produced (the worker's output content).
Inputs: taskId (required).
Get full task detail: status, proposals, approvals, executions, deliverables.
Inputs: taskId (required).
List tasks, optionally filtered by worker or status.
Inputs: limit, cursor, status, workerId.
Poll a task until it reaches a terminal or actionable status (PROPOSED, COMPLETED, FAILED, …) or the timeout. Returns the final task detail.
Inputs: taskId (required), timeoutMs.
List recent credit usage ledger entries (paginated).
Inputs: limit, cursor.
Subscribe a URL to CreateWorker events (task.completed, proposal.needs_approval, …). Returns a signing secret (once).
Inputs: url (required), events (required), description.
Create a worker. roleKey is a CreateWorker role (e.g. OPERATIONS_ASSISTANT, CUSTOMER_SUPPORT_AGENT, RESEARCH_ANALYST). Autonomy defaults per role.
Inputs: name (required), roleKey (required).
Archive (soft-delete) a worker.
Inputs: workerId (required).
Get one worker by id (role, status, autonomy level).
Inputs: workerId (required).
List the org's AI workers. Use a worker's id to create tasks for it.
Inputs: limit, cursor, status.
Update a worker's name and/or status.
Inputs: workerId (required), name, status.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Scored across 28 tools
Each tool targets a distinct resource and action (account, approval, binding, clarification, connection, deliverable, knowledge, task, usage, webhook, worker). The descriptions clearly differentiate between similar-sounding tools like task_deliverables vs deliverable_get and webhook_register vs connection_create_webhook. No overlapping purposes.
All tool names follow a consistent verb_noun pattern in snake_case (e.g., account_get, binding_create, task_cancel). Pluralization of nouns (bindings_list vs worker_get) is the only minor variation but does not detract from consistency.
28 tools cover a broad platform with workers, connections, tasks, knowledge, approvals, etc. While slightly on the higher side, each tool serves a clear purpose in the workflow and none feel redundant.
The tool set covers the full lifecycle for workers and tasks (CRUD, execution, approval, deliverables). Minor gaps include missing update functionality for knowledge documents and connections, but these are not critical for core workflows.
MCP server for your apps' tools and custom tools, plus hosted AI agents and approval-gated workflows
MCP server for AI agents to plan, verify, and deploy Cloudflare-native apps.
MCP server connecting AI agents to 100+ apps (Gmail, Slack, Notion, GitHub) via one-click OAuth.
MCP server that lets your AI assistant build and deploy apps, hosting each one at a live URL.
