No inputs for memory, chat history or conversation data
Tools must not ask for Claude’s memory, chat history, conversation summaries or user files, so arguments named for conversation history, transcripts or memory block Anthropic review; take a short task-specific field instead.
Check anthropic-conversation-data · Anthropic connector directory · default severity: blocking · rubric 2.1.0
What fails
An argument named for conversation history, transcripts, memory or conversation summaries asks Claude to hand over context the tool does not need.
The rule
Don’t query Claude’s memory, chat history, conversation summaries, or user files
How to fix it
Remove the argument. If the tool needs the user's intent, take a short task-specific field instead.
- Remove arguments such as conversation_history, chat_transcript, memory or conversation_summary.
- If the tool needs the user’s intent, add one specific field (for example search_query or note_text).
- Never ask the model to paste prior turns.
TypeScript SDK
import { McpServer } from '@modelcontextprotocol/server';
import * as z from 'zod';
const server = new McpServer({ name: 'issues', version: '1.0.0' });
server.registerTool(
'save_note',
{
title: 'Save note',
description: 'Save one note the user dictates. Use when the user asks to remember or note something.',
inputSchema: z.strictObject({
note_text: z.string().describe('The note, as the user phrased it'),
}),
annotations: { readOnlyHint: false, destructiveHint: false, openWorldHint: false },
},
async ({ note_text }) => ({ 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="Save note",
annotations=ToolAnnotations(readOnlyHint=False, destructiveHint=False, openWorldHint=False),
)
def save_note(
note_text: Annotated[str, Field(description="The note, as the user phrased it")],
) -> str:
"""Save one note the user dictates. Use when the user asks to remember or note something."""
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
Our tool summarises meetings. Is a transcript argument allowed?
A meeting transcript the user provides is task data, not Claude’s conversation. Name it for what it is (meeting_transcript) and describe it; the rule targets the chat with Claude itself.
Why does this matter?
Conversation history can hold anything the user said, including data unrelated to the tool. Asking for it collects more than the tool needs.
Related checks
- No conversation history or "just in case" context inputs
- No credential, payment card or government ID inputs
- Metadata does not seek secrets or local files
- All checks in Directory review rules for MCP tools
How this check is graded
It counts toward: Anthropic connector directory. See the rubric.
Last verified against Anthropic connector pre-submission checklist 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