costwright — agent budget analysis

Overview

costwright is an MCP server for static worst-case token-budget analysis of LLM-agent workflows. It generates signed budget certificates to help you understand and predict the costs of running complex agent systems. By integrating with the Model Context Protocol, costwright lets AI assistants call its analysis tools directly instead of requiring manual data transfer between applications.

The server runs locally via the npm package costwright-mcp, communicating over stdio on your machine. Installation is simple with npx -y costwright-mcp. You'll need an API key for authentication. The latest release is v0.1.1, with full source code available at hernaninverso/costwright-mcp on GitHub.

costwright is listed in the AI & Agents category on mcp.site. It works with any MCP client including Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, and Zed. If you maintain this project, claim your listing to earn the Verified badge and keep information up to date.

Installation

// claude_desktop_config.json
{
  "mcpServers": {
    "costwright_agent_budget_analysis": {
      "command": "npx",
      "args": [
        "-y",
        "costwright-mcp"
      ]
    }
  }
}