Agent Observability

Overview

Agent Observability is a Model Context Protocol (MCP) server: Trace AI agent execution: every tool call, every error, every dollar. Open source, local-first. In practice that means any MCP-compatible AI assistant can call Agent Observability's tools directly — the model decides when to use them in a conversation or agent run.

It runs locally: the npm package agent-obs speaks MCP over stdio on your machine. No credentials are required — it works out of the box. The current release is v1.0.7, published under the MIT license, with source at KryptosAI/agent-observability on GitHub.

Agent Observability is listed under AI & Agents on mcp.site and works with any MCP client — Claude Desktop, Claude Code, Cursor, VS Code, Windsurf, Zed and the rest of the ecosystem — using the install snippets below. If you maintain Agent Observability for KryptosAI, claim this listing to verify ownership, earn the Verified badge, and keep the details current.

Installation

// claude_desktop_config.json
{
  "mcpServers": {
    "agent_observability": {
      "command": "npx",
      "args": [
        "-y",
        "agent-obs"
      ]
    }
  }
}