ReferenceSlash-pickable prompts

Slash-pickable prompts

12 pre-canned prompts in Claude's prompt picker — type slash, pick, fill in any arguments, and the right Productised tool chain runs.

The connector ships 12 MCP Prompts — pre-canned templates and widget viewers users can pick from Claude's prompt menu instead of typing a sentence from scratch.

There are two kinds:

  • Parameterised templates (7) — pick the prompt, fill in arguments, and a formed message drops into the chat for Claude to act on
  • Widget viewers (5) — pick the prompt and the corresponding visual widget surfaces inline. No sentence needed, no sendMessage caution banner

Prompts are an additional access surface alongside the 80 tools — they don't replace anything, they just make common flows easier to discover and faster to invoke.


How users see them

In Claude.ai, click the + button in the chat input (or type /) and pick Productised to see the prompt list. Each prompt has a title, a description, and (if it takes arguments) an inline form to fill in.

In Claude Desktop, it's the same picker UI.


Parameterised templates (7)

These insert a formed prompt into the chat — like a slash command that becomes a real message.


Widget viewers (5)

These return an embedded resource link pointing to an existing visual widget. The Claude client renders the widget inline — no tool call, no sendMessage, no "external content" warning.


Why we ship both tools AND prompts

Tools are Claude-invoked. When the user says "build me a scorecard", Claude reads the request, picks the right tool, and runs it. Great for natural language; requires the user to phrase their intent clearly.

Prompts are user-invoked. The user picks from a menu. Great for discoverability (you don't have to know the right phrasing); requires us to anticipate the common flows.

The two complement each other. New users discover what's possible via the prompts menu; experienced users skip the menu and just talk to Claude.


A note on the "caution" banner

When a widget button in chat uses sendMessage to insert text into Claude's input, Claude shows a security warning: "Use caution before running this prompt. Malicious conversation content could trick Claude…". That's appropriate — it's protecting users from untrusted external content.

Widget viewers (the 5 view_* prompts) bypass that warning because the user explicitly picked the prompt from the official picker. Claude treats the invocation as deliberate user intent, not external content trying to send a message.

If you want a particular widget without the caution banner, use its viewer prompt instead of clicking a widget button that uses sendMessage.