Skip to content
IsoKron

destructiveHint: false on a write: the platforms word this differently

OpenAI says destructiveHint: false is right for additive writes with no destructive or irreversible effect; Anthropic asks for destructiveHint: true on tools that modify or delete data and does not say whether a pure create counts, so false is defensible only for writes that purely add.

Check anthropic-hint-platforms-differ · Anthropic connector directory · default severity: attention · rubric 2.1.0

What fails

This write tool sets destructiveHint: false. OpenAI says false is right for "additive writes without destructive or irreversible effects". Anthropic's checklist asks for destructiveHint: true on "tools that modify or delete data", and does not say whether a pure create counts as modifying. Neither text settles it, so this is not graded as blocking. If the tool can overwrite, cancel, revoke or send something irreversible, it is destructive on both platforms.

The rule

Every tool must include a title and the applicable hint: readOnlyHint: true for read-only tools, and destructiveHint: true for tools that modify or delete data.

Source: Anthropic connector pre-submission checklist

How to fix it

Keep destructiveHint: false only if the tool purely adds (a new record, a new comment) and cannot overwrite or delete. Otherwise set it to true. If a reviewer asks, the OpenAI wording supports false for additive writes.

  1. Keep destructiveHint: false only if the tool purely adds (a new record, a new comment) and cannot overwrite, cancel, revoke or send.
  2. If it can overwrite or has an irreversible effect, set destructiveHint: true.
  3. If a reviewer asks, cite OpenAI’s wording for additive writes.

TypeScript SDK

TypeScript
import { McpServer } from '@modelcontextprotocol/server';
import * as z from 'zod';

const server = new McpServer({ name: 'issues', version: '1.0.0' });

server.registerTool(
  'add_comment',
  {
    title: 'Add comment',
    description: 'Add a new comment to one issue; never edits or removes existing comments. Use when the user asks to reply on an issue.',
    inputSchema: z.strictObject({
      issue_id: z.string().describe('Issue ID, for example ISS-42'),
      comment_body: z.string().describe('Comment text in Markdown'),
    }),
    annotations: { readOnlyHint: false, destructiveHint: false, openWorldHint: false },
  },
  async ({ issue_id, comment_body }) => ({ content: [{ type: 'text', text: 'Done' }] }),
);

Python SDK

Python
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="Add comment",
    annotations=ToolAnnotations(readOnlyHint=False, destructiveHint=False, openWorldHint=False),
)
def add_comment(
    issue_id: Annotated[str, Field(description="Issue ID, for example ISS-42")],
    comment_body: Annotated[str, Field(description="Comment text in Markdown")],
) -> str:
    """Add a new comment to one issue; never edits or removes existing comments. Use when the user asks to reply on an issue."""
    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

Will Claude reject destructiveHint: false on a create tool?

The checklist does not say. That is why the checker marks it for attention, not as blocking. A tool that only adds data is the case OpenAI explicitly calls non-destructive.

What does destructiveHint change for the user?

Clients use it to decide whether to ask before a call. With false, an additive write may run with less friction; with true, the user confirms each call.

Further reading

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