mcp-cost-guard 🛡️

Zero-dependency CLI and Model Context Protocol (MCP) server for auditing LLM token costs, enforcing CI/CD prompt budgets, compressing context windows, and scrubbing leaked API keys before sending requests to Claude, OpenAI, or Gemini.

Features

• Instant Cost & Token Estimation: Supports claude-opus-4, claude-sonnet-4, claude-3-5-haiku, gpt-4o, and gemini-2.5-pro.

• CI/CD Budget Gatekeeper: Fail builds automatically (--max-budget 0.05) when oversized context files or system prompts exceed budget thresholds.

• Credential & Secret Scanner: Detects leaked Anthropic (sk-ant-...), OpenAI, GitHub, AWS, and private keys before prompts leave your machine.

• Native MCP Server: Plugs directly into Claude Code and Claude Desktop over stdio with zero third-party dependencies.

Related MCP server: ctxbleach

Quick Start

git clone https://github.com/Git-AnshGrover/mcp-cost-guard.git
cd mcp-cost-guard
pip install -e .

CLI Usage

# Audit a prompt file against Claude Sonnet 4
mcp-cost-guard --file system_prompt.md --model claude-sonnet-4

# Enforce a strict USD budget and secret scan in CI/CD
mcp-cost-guard --file context.txt --max-budget 0.10 --fail-on-secrets --json

# List all supported models and pricing per 1M tokens
mcp-cost-guard --list-models

Claude Code / Claude Desktop MCP Configuration

Add the following to your claude_desktop_config.json or .mcp.json:

{
  "mcpServers": {
    "cost-guard": {
      "command": "mcp-cost-guard",
      "args": ["--mcp-stdio"]
    }
  }
}

Contributing

We actively welcome first-time and experienced open-source contributors! Check our open Good First Issues and read CONTRIBUTING.md to get started in under 5 minutes.