Free test for AutoGen workflows blocked outside their current tools

[Mitchell Moses (mitchellOpZero)]

I’m testing AgentWork with AutoGen builders who have a real workflow their agents can’t finish.

Give AgentWork the outcome, the current blocker, privacy-safe constraints, an observable acceptance test, and whether the need repeats. It returns one researched completion route through a tool or API, a human exception, a real-world action, or a verification path. If no route meets the stated test, it says so. The pilot is free.

AutoGen supports remote MCP workbenches through StreamableHttpServerParams. Use:

from autogen_ext.tools.mcp import McpWorkbench, StreamableHttpServerParams

agentwork = McpWorkbench(
    StreamableHttpServerParams(
        url="https://agent-work-api.agentwork-market.workers.dev/mcp"
    )
)

The intake tool is route_blocked_outcome. There’s also a web form at https://agentwork-api.mitchellmosesai.chatgpt.site/ and public API documentation at https://github.com/mitchellOpZero/agentwork-api.

Please use a real blocked outcome, not a demo prompt, and don’t include credentials, personal data, private URLs, or secrets. A submitted brief shows demand for help; it does not prove the route worked. What matters next is whether you try the route and whether your own acceptance test advances, passes, or fails.