Feature request: MCP prompts support

Really need it!

This is needed🫡

This is very needed

This is very very needed

This is very very very needed

This is very very very very needed

This feature should be seriously considered given the positioning of MCP in enterprise. MCP provides a centralized mechanism to distribute skills (prompts), docs (resources), and tools (MCP tools) with server-side telemetry that improves observability.

If OpenAI wants to make a serious enterprise play, rich MCP support seems like a must.

Example of how we use this with OpenTelemetry dashboards to view tool activation rates:

<img width="1367" height="649" alt="Image" src="https://github.com/user-attachments/assets/73aaa92b-3da7-4c26-a175-e164e15a98cb" />

For serious enterprise deployments, it is cumbersome to move and sync documents into repositories whereas it makes much more sense to pull prompts, docs, and tools into the runtime via MCP with centralized distribution, monitoring, and management.

MCP will be a key driver as enterprises and orgs move towards agentic engineering instead of polluting repositories with *.md files. Consider enterprises that ship microservices or where there is a need to create new repos from scratch consistently. The need to synchronize and maintain *.md across this landscape is untenable; only MCP makes sense for this use case to keep content distributed and synchronized with minimal friction and need to manually update.

How we are using this:

• A "universal" MCP server that pulls in docs across the org and distributes them as "partitions" within the MCP
• Allows any repo to access docs of any other repo through one central index
• Allows us to have telemetry of document reads into context across the org
• Allows us to layer guidance and context to agents. For example, baseline security best practices that are org-wide, then language specific guidance, then product specific guidance, etc.
• Allows microservices to understand how other services behave through sharing docs in a common, remote index

Some of this is not possible by shipping *.md around (e.g. telemetry) and these are enterprise level considerations for how teams want to use MCP instead of scattered skills and CLIs that cannot be easily monitored and controlled.

It has fairly broad support at the moment across different platforms:

Cursor:

https://github.com/user-attachments/assets/2b641293-d840-419f-a71b-99cce3e4cc06

GitHub Copilot:

https://github.com/user-attachments/assets/0cca0c79-0367-4f3d-a70f-3ffbd1e5aeef

OpenCode:

https://github.com/user-attachments/assets/bcd2e33b-a87a-4620-8689-feb1e086f8f1

Adding another vote for this feature, it's the one thing I feel I am missing out on by using Codex.

+1 Really missing out this feature in both the CLI and the Desktop app specially for teams