We’re adding support for MCP Tasks and Elicitations in Claude. Tasks let Claude give plugins long-running jobs such as deep research and report back when they’re done.

Elicitations let plugins ask users for input mid-task, such as confirming a refund or filling in a missing detail.

Tasks and Elicitations are available today in the Claude Code CLI, rolling out to more Claude products soon. Building plugins for Claude? Add Tasks and Elicitations to your MCP server today.

That’s great but fix your nerfed usage limits that started this week

Fantastic... but can you fix opus to do what we tell it to do and not lie about it? Every day this week the retarded faggot opus 5.5 max model lied about completing work Is there a Hebrew setting that we can disable?

definitely designed by Claude

I can't even keep up with the updates 😭 Its time to become unemployed

the no call held open label in that screenshot is the real feature. nothing sits there waiting while the job runs

Someone on the dev team is quite the Sesquipedalian.

This is incredible. There are already MCP and Plugin but Dev build persistence of these functions. INCREDIBLE! 4 days 4 giant news tgese week!!! Please: pumping Claude with MCP for Unreal, because we need a worker inside Unreal. Please make Claude image to Blender 1:1

whoa, jobs that report back on their own. my long runs still get a nudge from me every few minutes, would love to see how far tasks get without that

Tasks has been in the MCP spec as experimental since Nov 2025. Good to see it finally land in a first-party client, since long-running research was the missing piece for real background agents.

we can finally stop building custom polling loops inside mcp servers. saves a lot of plumbing.

That’s great… when are you gonna fix the excessive weekly limit drainage problem?

Elicitations fit refunds and deletes best. Show the amount and ask a yes/no, never a free-text "confirm".

Good one for builders learning MCP. Tasks handle the long jobs, Elicitations keep a human in the loop when it matters. Practical building blocks for real-world AI tools

Just noticed this today! Though Atlassian MCP tasks always fail (fault on them).

Please fix the usage quota problems

mcp tasks are poll-first: the client calls tasks/get on the server's interval until a terminal state, and push notifications are optional. so 'report back' is a client loop. input_required is the interesting one: a job that stops to ask waits on the client. who answers overnight?

what happens to an elicitation in claude -p? fails fast, or waits for an answer nobody is there to give

Finally an official way for the agent to say "wait, are you sure?" before it refunds the whole order 😂

elicitations is the one i’m most excited about. a plugin that can stop and ask instead of guessing is a much safer plugin

Finally. Right now my agents report back by pinging my phone through ntfy Tasks means the job reports back on its own. Elicitations means it asks me before guessing The guessing is where all my bugs come from

tasks is the one i've been waiting for. half my mcp tools were faking async with polling loops and timeouts elicitations too, finally a clean way for a server to ask "which account?" mid-run instead of guessing

@grok make the all the way down meme but its mcp

Duet eta

@bot explain me this new feature. Check this entire thread

Elicitations are the piece I wanted most: the plugin can stop mid-task and ask instead of guessing. Curious about the timeout case. If nobody answers, does the task wait, or fall back to a safe default?