• SAFE
  • MUTATION
  • RISKY
  • HUMAN

How to verify tool side effect verbs with your AI

Connect this MCP to your agent to audit your tool definitions. It catches missing mutation flags before your AI executes a command that changes data.

Ask AI about this page

Short answer

How do I make sure my AI asks before running destructive tools?

Run a validation check on your tool manifest to ensure every action that modifies data is explicitly marked as a mutation. This forces your agent to trigger confirmation prompts instead of running commands blindly. You'll get a clear report of which tools are safe and which ones need permission flags.

The audit results.

Where every issue lands.

The AI categorizes your tool definitions into these four states.

SAFE

Read-only tools

The AI confirms these tools only fetch data. No confirmation prompts are needed for these actions.

MUTATION

Correctly flagged

These tools are marked as mutations. Your agent will now ask you for permission before running them.

RISKY

Missing mutation flags

The AI found destructive tools that aren't marked as mutations. These are the ones that could cause issues.

HUMAN

Manual review needed

The AI flags ambiguous tool names that require you to manually decide if they should be mutations.

The workflow

What your AI does when the manifest arrives.

The agent parses your tool definitions to find discrepancies between action names and their mutation status.

  1. Scan manifest

    The agent reads your tool definitions to map out every available action and its current configuration.

    validate_mcpfusion_implementation

  2. Identify mutations

    The AI looks for verbs that imply data changes, like delete, update, or create, and checks their flags.

    validate_mcpfusion_implementation

  3. Flag mismatches

    It isolates tools that perform destructive work but lack the necessary mutation metadata.

    validate_mcpfusion_implementation

  4. Generate report

    The agent presents a list of safe tools and a list of high-risk tools that need your attention.

    validate_mcpfusion_implementation

Try it

Copy these to start.

Use these prompts to begin your audit.

Starting points

Replace the specific tool or file names with your own implementation details.

MCPFusion Developer Prover Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_kq18iP2KBas7gK6lLEfjXD7wkf1LFvR2zOIJQ4BQ/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — MCPFusion Developer Prover capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "mcpfusion-developer-prover-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_kq18iP2KBas7gK6lLEfjXD7wkf1LFvR2zOIJQ4BQ/mcp"
    }
  }
}
Copy into chat04

  • Validate my current tool manifest and tell me which ones are missing mutation flags.

  • Check the 'update_user_profile' tool in my manifest to see if it's correctly marked as a mutation.

  • Review my MCP implementation and list all tools that could perform destructive actions without permission.

  • Scan my tool definitions and identify any read-only tools that are accidentally marked as mutations.

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect MCPFusion Developer Prover once, then ask.

Just link the MCP to your client. Your credentials stay encrypted, and you can start auditing your tools immediately.

Connect MCPFusion Developer Prover to your AI

FAQ

How this task behaves.

  • 01

    Will the AI automatically fix my tool manifest?

    No. The AI only identifies the errors and flags them for you. You must manually update your code to add the mutation flags.

  • 02

    Can the AI delete tools that are missing flags?

    No. The AI has no permission to modify your files or delete any tools. It only reads and reports.

  • 03

    How does the AI know a tool is destructive?

    It analyzes the tool name and the provided schema to see if the action modifies, creates, or deletes data.

  • 04

    Does this work with any MCP-compatible client?

    Yes, as long as your client supports the Model Context Protocol, you can use this to audit your tools.

  • 05

    Can I use this to check my custom Python tools?

    Yes, if your tools are exposed through an MCP-compliant manifest, the AI can validate them.

  • More questions about MCPFusion Developer Prover? The Connector page answers them. See everything the MCPFusion Developer Prover Connector can do

Connect MCPFusion Developer Prover to Claude, Cursor, ChatGPT & more