Expose MCP Server Prompts as Slash Commands Like Claude Code

Resolved 💬 6 comments Opened Dec 19, 2025 by baron Closed Aug 6, 2026

What feature would you like to see?

The MCP protocol defines MCP server prompts as "a standardized way for servers to expose prompt templates to clients". This is currently well supported by Claude Code and VS Code.

This allows MCP Servers to:

Provide well-maintained and up-to-date custom prompts to users
Free users from manually checking and refreshing available prompts from MCP providers
Live updates to prompts without restarts
Better discoverability inside the existing Codex slash popup
Structured parameters, fewer “how do I call this?” failures
Better security than copy-pasted prompts
Respects company policies that do not allow custom prompts
Ecosystem parity and portability

This makes it extremely convenient to invoke tools like RepoPrompt by @provencher which involve sophisticated workflows subject to frequent updates.

  /RepoPrompt:rp-build (MCP)           Build with RepoPrompt MCP context_builder → chat → implement. A structured workflow for implementing features using deep codebase context.
                                       (arguments: task)
  /RepoPrompt:rp-investigate (MCP)     Deep bug investigation with RepoPrompt MCP tools. A systematic protocol for investigating issues using context_builder and structured analysis.
                                       (arguments: issue)

Codex already supports MCP servers and a keyboard-first command palette; adding MCP prompt discovery (prompts/list) + execution (prompts/get) as slash commands unlocks centrally managed, updatable, parameterized workflows with less user friction than local prompt files and better governance than copy-paste.

Additional information

_No response_

View original on GitHub ↗

6 Comments

w-winter · 8 months ago

baron · 7 months ago

@tibo-openai sir please

eloquence · 7 months ago

Yes, this would be very nice to see. MCP prompts are quite powerful for serious multi-step workflows.

+1 to the /server:prompt format, which seems to be emerging as the de-facto standard. Maybe disambiguate with a shortened hash of the server name if it's conflicting with a built-in command (/server-a28b:prompt etc.)

hmcezar · 4 months ago

This would be a great feature indeed.

szybnev · 3 months ago

+1. Exposing MCP prompts as slash commands is one of the highest-value Claude Code parity features for explicit workflows. Skills are useful when the agent decides they apply, but MCP prompts as /server:prompt give the user a deterministic, discoverable way to invoke a maintained workflow with structured args without copy-pasting prompt text.

This would also make MCP-provided workflows more portable across clients and easier for teams to standardize.

m13v · 1 month ago

the policy argument holds at prompts/list, where an admin picks which servers publish templates. it stops at prompts/get: the expanded prompt reaches the tool layer as an ordinary turn, unmarked. the thing that blocks copy-paste sees the same call.