MCP Command and Integration Guide
How to operate MCP servers and tools safely from Codex.
Key takeaways
- The core MCP-related commands are
/mcp,/apps,/plugins, and/permissions. - Start integrations with read-only tools before granting broader access.
- Design the approval policy at the same time you design the tool surface, not afterward.
- Document fallback steps for server failures, auth failures, and missing tools.
- For design tools such as Figma, MCP supplies external context while a Skill defines the retrieval, implementation, and verification sequence.
Core Commands
/mcp/apps/plugins/permissions
Connect design tools
In the official Codex design workflow, the Figma MCP server supplies structured design context, variables, assets, and the exact variant. Check the following before implementation:
/plugins
/skills
/mcp| Check | Purpose |
|---|---|
get_design_context | Pull structure, styles, and asset context for the exact Figma node |
get_metadata | Map the file and narrow the node set when a response is truncated |
get_screenshot | Capture the exact visual variant before coding |
$playwright | Verify responsive behavior, interactions, and visual differences in a real browser |
If tools are missing, use /mcp verbose to inspect authentication and tool exposure. Follow /en/books/codex-command-master/design-workflows for the full implementation loop.
Operating Principles
- Start with read-only tools.
- Design the approval policy at the same time as the tool surface.
- Document fallback steps for server failures, auth failures, and missing tools.
- Record the design source of truth and final browser viewports.
Sources
- Codex MCP customization: https://learn.chatgpt.com/docs/customization/overview#mcp
- Figma designs to code: https://learn.chatgpt.com/use-cases/figma-designs-to-code