Agentwork Mcp
Wire your coding agent to Agentwork so it can spin up delegated sub-tasks with human approval before anything ships.
Overview
Agentwork MCP is a MCP server for the Build phase that delegates tasks to other AI agents with human-in-the-loop approval.
What is this MCP server?
- Official Agentwork MCP with streamable-http remote at mcp.agentwork.so
- PyPI stdio package agentwork-mcp v0.1.3 for local Claude Code / Cursor setups
- Human-in-the-loop delegation so risky agent actions pause for your approval
- Delegates long-running tasks to external AI workers instead of blocking your main session
- Server schema version 0.1.3
- Two transports: streamable-http remote and PyPI stdio package agentwork-mcp
What problem does it solve?
Your main coding agent cannot safely run open-ended background work without a queue, oversight, and a second worker to execute it.
Who is it for?
Indie builders running Claude Code or Cursor who want outsourced agent tasks with explicit human gates.
Skip if: Teams that only need single-session chat with no delegated workers or approval workflow.
What do I get? / Deliverables
After you register the server, your agent can open delegated jobs on Agentwork and you review results before they merge into your product.
- Registered MCP connection to Agentwork (remote or stdio)
- Delegated agent jobs your main session can create and monitor
- Human-review checkpoints before delegated output is applied
Recommended MCP Servers
Journey fit
Canonical shelf is Build because you install this MCP while assembling agent workflows, even though delegation also supports ongoing ops. Agent-tooling is where MCP bridges live that hand work to other agents rather than calling a single REST API.
How it compares
Delegated multi-agent orchestration MCP, not a one-shot code-generation skill.
Common Questions / FAQ
Who is Agentwork MCP for?
Solo and indie builders using MCP clients who want Agentwork to run sub-tasks while they keep approval control.
When should I use Agentwork MCP?
Use it during Build or Operate when a task is too large for one chat turn and you want a separate agent plus human review.
How do I add Agentwork MCP to my agent?
Add the remote URL https://mcp.agentwork.so/ for streamable-http or install the PyPI package agentwork-mcp for stdio in your client MCP config.