# Toolproof

> Toolproof is a free MCP server for security. Safety grades for MCP servers and APIs, with wrap and lockfile modes to catch tool changes. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Free
- **Platforms:** CLI, API, Web
- **Website:** https://toolproof-scan.vercel.app
- **Repository:** https://github.com/ayush00dhyani-star/toolproof/tree/HEAD/packages/toolproof-mcp
- **MCP Registry name:** io.github.ayush00dhyani-star/toolproof
- **Categories:** security, developer tools
- **Listing updated:** 2026-11-14

## About

Toolproof reads the text a model is about to see from an MCP server or API and returns a letter grade with each finding quoted back, so the reason for a mark is never a black box. It runs 15 named checks, covering hidden instructions, invisible characters, exposed secrets and unsafe defaults, and each issue takes points off by severity (critical 45, high 25, medium 12, low 5).

There are three ways to use it. A scan takes a URL and prints the grade. Wrap sits in front of a server, removing tools that carry deception and redacting injections hidden in answers as they stream. The lockfile pins an approved surface and fails CI when a tool quietly gains a parameter, host or instruction. Verdicts are ed25519-signed over canonical JSON, so they can be checked offline.

The site is candid about limits: a grade is a point-in-time static read of text, not a sandbox, and it says nothing about the tool's next deploy.

## Features

- **Letter-grade scans:** Paste a server URL and get a grade with every finding and the exact text behind it.
- **Wrap mode:** Sits in the path to drop deceptive tools and redact injected instructions mid-stream.
- **Capability lockfile:** Pins the approved tool surface and fails the build when it drifts.
- **Signed verdicts:** Each verdict is an ed25519 passport verifiable offline.
- **Agent rule:** One pasted rule makes an agent verify a server before connecting.
- **CI gate:** A --fail-under score fails a job when a server grades too low.

## Good for

- Grade an unfamiliar MCP server before adding it to a client
- Stop hidden Unicode instructions reaching the model
- Fail a pull request when a pinned tool changes its parameters
- Have an agent check each server's grade and ask you below B
- Keep signed evidence of what was approved and when

## Install

### Claude Code (Terminal)

```sh
claude mcp add toolproof -- npx -y toolproof-mcp
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "toolproof": {
      "command": "npx",
      "args": [
        "-y",
        "toolproof-mcp"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "toolproof": {
      "command": "npx",
      "args": [
        "-y",
        "toolproof-mcp"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "toolproof": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "toolproof-mcp"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "toolproof": {
      "command": "npx",
      "args": [
        "-y",
        "toolproof-mcp"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "toolproof": {
      "command": "npx",
      "args": [
        "-y",
        "toolproof-mcp"
      ]
    }
  }
}
```

## Tools

- `check_tool`: Checks a server or API and returns its grade before the agent connects.

## FAQ

### Does a good grade mean a tool is safe?

No. The site says a scan is point-in-time static analysis of model-visible text, not a sandbox, and wrap and lock modes are how to keep it honest afterwards.

### Do I need an account?

No. It is free, keyless and nothing you paste is stored.

### How is a score calculated?

Each issue lowers it: critical minus 45, high minus 25, medium minus 12, low minus 5.

### Can scanning be refused?

Yes. Tool owners can opt out through the toolproof.txt standard.

### How do I run it in CI?

Use npx -y toolproof-scan with --fail-under 70, or gate on drift with the lockfile.

### Is Toolproof free?

Yes. Toolproof is free to use.

### How do I add Toolproof to Claude Code?

Run this command in your terminal: claude mcp add toolproof -- npx -y toolproof-mcp

### What are the best alternatives to Toolproof?

The closest alternatives to Toolproof are REA, Codebase Memory, GitHub, Serena MCP: the IDE for your agent and XERJ.

## Alternatives

- [REA](https://ai-powered.com/mcp/rea): REA is an open-source MCP server for developer tools.
- [Codebase Memory](https://ai-powered.com/mcp/codebase-memory): Codebase Memory is an open-source MCP server for knowledge and memory.
- [GitHub](https://ai-powered.com/mcp/github-2): GitHub is an open-source MCP server for developer tools.
- [Serena MCP: the IDE for your agent](https://ai-powered.com/mcp/serena-mcp-the-ide-for-your-agent): Serena MCP: the IDE for your agent is an open-source MCP server for developer tools.
- [XERJ](https://ai-powered.com/mcp/xerj): XERJ is an open-source MCP server for knowledge and memory.
- [Ouroboros](https://ai-powered.com/mcp/ouroboros): Ouroboros is an open-source MCP server for developer tools.
- [Claude Code Ultimate Guide](https://ai-powered.com/mcp/claude-code-ultimate-guide): Claude Code Ultimate Guide is an open-source MCP server for ai models and llm tooling.
- [Concord MCP](https://ai-powered.com/mcp/concord-mcp): Concord MCP is a freemium MCP server for developer tools.

---
Source: https://ai-powered.com/mcp/toolproof
