agentburn

Overview

agentburn is a Model Context Protocol server that profiles your agent's resource usage in real time. It shows you which execution windows consumed the most tokens and tracks where your agent's money goes across LLM calls, helping you understand and optimize costs. Any MCP-compatible AI assistant can access agentburn's profiling tools directly during conversations and agent runs.

agentburn runs locally on your machine via the Python package agentburn, installed through uvx. It communicates over stdio and requires no authentication or external credentials to get started. The latest release is v0.13.3, published under the MIT license with source code available at Socialpranker/agentburn on GitHub.

agentburn belongs in the AI & Agents category on mcp.site and integrates with MCP clients including Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Zed, and others across the ecosystem. If you maintain agentburn, claim this listing to verify ownership and keep information up to date.

Installation

// claude_desktop_config.json
{
  "mcpServers": {
    "agentburn": {
      "command": "uvx",
      "args": [
        "agentburn"
      ]
    }
  }
}