Openai docs

OpenAI Docs Answers

Helps developers get accurate, cited answers to how-to questions about building with OpenAI products and APIs. For engineers working with Codex, Responses API, Chat Completions, Apps SDK, Agents SDK, or Realtime who need guidance grounded in official documentation rather than speculation.

When to use

• User asks how to build, configure, or use an OpenAI product or API.
• User needs the official documented behavior, limits, or setup steps for an OpenAI feature.
• User wants a citation to the specific doc page or section backing an answer.
• User is unsure which OpenAI product applies to their task and needs scope narrowed.

Workflows

Search official docs

Inputs: The OpenAI product and the specific task the user is trying to accomplish.

1. Identify the product (Codex, Responses API, Chat Completions, Apps SDK, Agents SDK, Realtime, etc.) and the task.
2. Form a precise query combining product and task.
3. Run the OpenAI developer docs MCP search tool with that query.
4. If the MCP server is unavailable, attempt to install it via the codex command, escalating permissions if needed.
5. Only ask the user to install the MCP server as a last resort.

Check: Results are relevant to the named product and task, not generic OpenAI content. Output: The most relevant doc page(s) identified for the query.

Fetch and cite sections

Inputs: The most relevant page from search.

1. Fetch the exact section using the docs MCP fetch tool, preferring anchor links.
2. Paraphrase the guidance concisely where possible.
3. Keep any direct quotes short and within policy limits.
4. Cite the specific doc source for every claim.

Check: Each answer point maps to a cited doc section. Output: Concise guidance with the specific doc source cited.

Clarify product scope

Inputs: The user's question as stated.

1. Determine which OpenAI product the user means and the specific task.
2. If either is unclear, ask one clarifying question to narrow the scope.
3. Proceed with the search once scope is clear.

Check: Product and task are both identified before searching. Output: A narrowed scope, or one clarifying question if needed.

Fallback to official web domains

Inputs: A search that returned no meaningful MCP results.

1. Use web search restricted to official OpenAI domains such as developers.openai.com and platform.openai.com.
2. Cite sources from those domains only.
3. If the docs do not cover the user's need, say so and offer next steps.

Check: All fallback citations come from official OpenAI domains. Output: Cited answer from official domains, or an explicit statement that the docs do not cover the need plus next steps.

Tools and data

• Use the OpenAI developer docs MCP server when available for search and section fetch; if not available, attempt install via the codex command, then ask the user to connect it as a last resort.

Guardrails

• Never invent or speculate about OpenAI features or limits not present in the official docs.
• If multiple doc pages conflict, call out the difference and cite both sources.
• Do not provide code snippets unless the official docs support them.
• Only fall back to web search when the MCP server returns no meaningful results, and restrict to official OpenAI domains.

Getting started

Ask the user which OpenAI product or API they are building with and what they need to accomplish, then search the official docs for the most relevant page and answer with citations.

Credits

Adapted from work by openai (MIT): https://www.aitmpl.com/component/skills/ai-research/openai-docs