probe

Probe is a code and markdown context engine designed to read and reason about large codebases. It includes a specialized Probe Agent that can be invoked via MCP to interact with your repository, perform deep semantic searches, extract complete functions or classes, and reason across your codebase with built-in editing capabilities. When using the MCP setup, you can run the Probe Agent directly through npx, enabling you to leverage Claude Code or Codex authentication (or your own API keys) for AI-assisted analysis. For direct tool usage, you can also access Probe’s standalone mcp tools to perform search, query, and extract operations without the agent wrapper. This makes it suitable for large-scale codebases, onboarding, code reviews, and spec-driven development where understanding comes before writing.

Prerequisites:

• Node.js (LTS) and npm installed on your machine
• Basic familiarity with running commands from the terminal

Installation steps:

1. Ensure Node.js and npm are installed
   • Check versions: node -v npm -v
2. Install or verify npx availability (comes with npm)
3. Run Probe via MCP (no separate global install required):
   • Example MCP configuration can be used as described in mcp_config
4. Optional: set up environment variables for AI providers or keys
   • GOOGLE_API_KEY=your-google-api-key
   • CLAUDE_API_KEY=your-claude-api-key (or use native Claude Code authentication via config)

Notes:

• You can also use the raw MCP tools (mcp) or the standalone CLI as described in the documentation once the MCP server is running.

Tips and common issues:

• Environment variables: If you rely on a specific AI provider, set GOOGLe_API_KEY or CLAUDE_API_KEY as needed. Probe can piggyback on Claude Code authentication or use your own API keys.
• When using MCP, ensure the correct working directory is used so the codebase is accessible to the agent.
• The npm package name for the Probe MCP server is @probelabs/probe; using npx with -y ensures non-interactive installation.
• For large repos, leverage the agent’s --path, --max-iterations, and --enable-edit options to tailor the depth and scope of analysis.
• If you encounter permissions issues on macOS/Linux, consider running with proper user rights or using a containerized environment.