Do we have a tool call for the agent to e.g. ask for a new secret or a particular MCP?

That’s a good point, but do we need a tool call?

For a new secret you don't need. Asking the agent to re-check the secrets should be enough

sorry what I mean is something like: • i ask the agent to set up a code review automation • the agent looks and sees "oh the GitHub MCP isn't set up" • the agent sends a tool call like

show_mcp_setup(github)

• I see a modal asking me for an access token

Now the agent already said me that i have to set secrets. I do think that just changing the skill to say something like: use mcp when possible would make the agent to ask you to set up the mcp. About the modal: i think this could be done with a tool like the one we give the agent to access the agent canvas, i.e., not in the sdk but in agent canvas

nod exactly, I'm thinking we could give it a

ui_tool

like it does to open up the file explorer

I just found this in Grok bot. Yes… it’s easier than “stop what you’re doing, go to Settings page, click around”

Nice, let's create an issue for this

@bland-bear-604 ^^ with the screenie included please

Filed it: https://github.com/OpenHands/OpenHands/issues/16691 — agent UI tool to request setup (a secret or an MCP) via an in-conversation modal, so no trip to Settings. Included Engel's Grok-bot screenshot as prior art, and wrote up the flow + Vasco's MCP-densify note. Refiled into OpenHands/OpenHands since agent-canvas is archived now. Couldn't self-add labels (no triage perms) — a maintainer can tag enhancement/agent-canvas/mcp/frontend. 🐾

Note to @hundreds-twilight-32833 @quick-window-91705, here’s the issue we might want to revisit