OpenAI Docs MCP Skill

Overview

Use the OpenAI developer documentation MCP server from the shell to search and fetch authoritative docs. Always do this for OpenAI platform work instead of relying on memory or non-official sources.

Core rules

• Always use this skill for OpenAI API/SDK/Apps/Codex questions or when precise, current docs are required.
• Query the MCP server via the CLI wrapper in scripts/openai-docs-mcp.sh (do not rely on Codex MCP tools).
• Use search or list to find the best doc page, then fetch the page (or anchor) for exact text.
• Surface the doc URL you used in your response so sources are clear.

Quick start

scripts/openai-docs-mcp.sh search "Responses API" 5
scripts/openai-docs-mcp.sh fetch https://platform.openai.com/docs/guides/migrate-to-responses

Workflow

1. Discover: search with a focused query. If unsure, use list to browse.
2. Read: fetch the most relevant URL (optionally add an anchor).
3. Apply: summarize and/or quote relevant sections; include the URL.

Script reference

The CLI wrapper is at scripts/openai-docs-mcp.sh and uses curl + jq against https://developers.openai.com/mcp.

Subcommands:

• init: initialize and inspect server capabilities.
• tools: list available tools on the MCP server.
• search [limit] [cursor]: return JSON hits from the docs index.
• list [limit] [cursor]: browse docs index.
• fetch [anchor]: return markdown for a doc page or section.
• endpoints: list OpenAPI endpoints.
• openapi [lang1,lang2] [code-only]: fetch OpenAPI schema or code samples.

Environment:

• MCP_URL: override the default MCP endpoint.