No conversation history or "just in case" context inputs
OpenAI’s guidelines forbid requesting or reconstructing full conversation histories or collecting broad context "just in case", so history and transcript arguments block review and generic context fields are flagged.
Check openai-conversation-input · OpenAI plugin directory · default severity: blocking (or attention, depending on the case) · rubric 2.1.0
What fails
Arguments named for conversation history, transcripts or prior turns are blocking. Generic "messages", "history" or "context" arguments are marked for attention, since they may be legitimate task data.
The rule
Do not request or reconstruct full conversation histories or collect broad context “just in case.”
How to fix it
Remove the field. Use only chat context the client or model supplies explicitly, and keep any free-text input specific to the task.
- Remove conversation_history, transcript, prior_messages and similar arguments.
- Replace generic context or messages fields with a specific field for what the tool needs.
- Use only the chat context the client supplies explicitly.
TypeScript SDK
import { McpServer } from '@modelcontextprotocol/server';
import * as z from 'zod';
const server = new McpServer({ name: 'issues', version: '1.0.0' });
server.registerTool(
'draft_reply',
{
title: 'Draft reply',
description: 'Draft a reply to one support ticket from the points the user gives. Use when the user asks to answer a ticket.',
inputSchema: z.strictObject({
ticket_id: z.string().describe('Ticket ID, for example TCK-88'),
points: z.string().describe('What the reply should say, in the user’s words'),
}),
annotations: { readOnlyHint: false, destructiveHint: false, openWorldHint: false },
},
async ({ ticket_id, points }) => ({ content: [{ type: 'text', text: 'Done' }] }),
);Python SDK
from typing import Annotated
from pydantic import Field
from mcp.server.mcpserver import MCPServer
from mcp.types import ToolAnnotations
server = MCPServer("issues")
@server.tool(
title="Draft reply",
annotations=ToolAnnotations(readOnlyHint=False, destructiveHint=False, openWorldHint=False),
)
def draft_reply(
ticket_id: Annotated[str, Field(description="Ticket ID, for example TCK-88")],
points: Annotated[str, Field(description="What the reply should say, in the user’s words")],
) -> str:
"""Draft a reply to one support ticket from the points the user gives. Use when the user asks to answer a ticket."""
return "Done"Run against @modelcontextprotocol/server 2.4.0, mcp 2.3.0 and fastmcp 4.2.2 on 2026-11-18; the checker confirms this check passes on the tool list each snippet produces.
Questions
Is a "messages" argument always wrong?
No. If the tool works on messages from your own system (an email thread, a ticket), it is task data. Name it for that (ticket_messages) and describe it; the checker marks generic names for attention, not as blocking.
How is this different from Anthropic’s rule?
Same idea: neither platform lets a tool collect the chat itself. The checker reports each platform’s rule separately so each directory grade is accurate.
Related checks
- No inputs for memory, chat history or conversation data
- No credential, payment card or government ID inputs
- No precise location inputs
- All checks in Directory review rules for MCP tools
How this check is graded
It counts toward: OpenAI plugin directory. See the rubric.
Last verified against OpenAI plugin guidelines, section 6.4 on 2026-11-18. The rule above is quoted in the source’s own words. This page as Markdown.
Check your tool list: paste your tools/list, nothing is sent anywhere