[FEATURE]:background task notification when long-running tool calls complete #36230

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

When using MCP tools that run async/background jobs (common with AI orchestration servers, CI pipelines,
build tools, etc.), the agent has no way to know when the job completes other than polling a status endpoint
repeatedly.

A typical 2-4 minute background job results in 20-40 redundant polling calls that:

• Waste tokens and clutter the conversation
• Make the agent appear stuck/frozen to the user
• Prevent the agent from doing other productive work while waiting
• Provide zero value until the final poll returns "finished"

Expected Behavior

The agent should be able to dispatch a background tool call and receive a notification when it completes,
rather than manually polling in a loop.

Possible approaches:
Expected Behavior

The agent should be able to dispatch a background tool call and receive a notification when it completes,
rather than manually polling in a loop.

Possible approaches:

1. Consume MCP notifications/progress — The MCP spec already defines progress/task notifications for this
   exact purpose. If opencode listened for server-pushed notifications and injected them into the agent
   context, polling would be unnecessary.

2. Background tool call mode — Allow marking a tool call as async, where the result is delivered later (
   pushed into conversation when ready). Agent continues other work in the meantime.

3. CLI progress indicator — Show a spinner/status in the terminal while waiting, and alert the agent/user on
   completion.
   Current Behavior

Agent must call a get_status(jobId) tool every 5 seconds in a loop:

status_check → "running" (useless)
status_check → "running" (useless)
status_check → "running" (useless)
... × 20
status_check → "finished" (finally useful)

This is the equivalent of while true; do curl status; sleep 5; done — functional but wasteful.

Use Cases

• MCP servers that dispatch long-running LLM/AI jobs
• Build/test pipelines triggered via MCP tools
• Any async workflow where result takes >30 seconds