[FEATURE] Claude Code CLI MCP Prompts do not elicit user input for prompt arguments

[CharlieDigital]

Preflight Checklist

• I have searched existing requests and this feature hasn't been requested yet
• This is a single feature request (not multiple features)

Problem Statement

When an MCP server exposes Prompts (https://modelcontextprotocol.io/specification/2025-03-26/server/prompts) with required arguments/parameters, Claude Code does not ask the user for inputs for these required parameters. This causes the prompt call to time out and fail.

Proposed Solution

When CC uses an MCP prompt,

• It should read the arguments/parameters and ask the user for values for those parameters.
• It should show the descriptions for those parameters

Alternative Solutions

The workaround is to explicitly provide the parameters when invoking the tool; however, users may not be aware of the presence of the parameters at all since the user interface does not provide clues that there are parameters required.

Priority

Medium - Would be very helpful

Feature Category

CLI commands and flags

Use Case Example

1. Provide an MCP server which advertises standard enterprise-wide pre-configured prompts
2. Prompt has one or more required parameters
3. User adds the MCP server and sees the prompts when using / command
4. User attempts to use the prompt without providing any parameters
5. The MCP server call fails because user does not provide required parameters

Additional Context

Kapture.2026-03-04.at.09.35.25.mp4

The MCP invoke times out from the missing parameters and fails.

Here is how VS Code GH Copilot handles the same interaction by eliciting the values for the required parameters from the user:

Kapture.2026-03-04.at.09.39.52.mp4

[github-actions]

Found 3 possible duplicate issues:

1. [BUG] Slash command picker auto-submits commands that require arguments (should insert into input) #23781
2. [FEATURE] Selecting slash command from autocomplete should insert text, not execute immediately #25500
3. Enhance MCP prompt argument handling with multi-word support and interactive discovery #4125

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

[CharlieDigital]

This is related to #23781, but is specific to MCP tools and the CLI

[ctolmsted]

Server-side verification: the MCP SDK plumbing is already there

I've been building an MCP server with prompt completions and traced through the bundled cli.js (v2.1.84) to understand where this breaks down. Summary:

What works (server → SDK):

• The embedded @modelcontextprotocol/sdk Client class has a fully functional complete() method for the completion/complete JSON-RPC call
• Schema validation exists for both request and response
• Capability assertion (serverCapabilities?.completions) is wired up
• Both ref/prompt (prompt argument completion) and ref/resource (resource URI completion) reference types are supported

What's missing (SDK → UI):

• client.complete() is never called anywhere in the application code — zero invocations in the entire bundle
• MCP prompts register as slash commands with static argument hints (e.g. [topic] [style] via the T34 hint renderer), but no dynamic completion is ever fetched
• The input handler (Xp4) already detects when the user is in the argument position of an MCP prompt command and sets commandArgumentHint — this is the precise insertion point for fetching completions

Suggested implementation path:

1. In the argument-position detection branch of the input change handler (M6), add a debounced call to client.complete() with ref: {type: "ref/prompt", name: promptName} and argument: {name: argName, value: partialValue}
2. Gate on server.capabilities?.completions — fall back to existing static hints for servers that don't support it
3. Surface results through the existing suggestion dropdown (same pattern as file suggestions via @)
4. For required arguments without completions support, prompt the user interactively (the core ask of this issue)

The gap is narrow — the SDK has the method, the UI has the suggestion infrastructure, they just aren't connected.