Terminal Agents

Connect command-line workflows like Claude Code directly to your index.

Terminal-based AI agents operate headless, but still require structural intelligence when modifying your project. Connecting SourcePrep locally allows them to execute semantic searches and assess blast radiuses without leaving the command line.

Integration Setup

Select your Terminal AI agent below. Update the respective configuration file with the provided JSON block to register SourcePrep as an MCP tool. Ensure the SourcePrep daemon is running (prep serve) before starting a chat session in the terminal.

Claude Code

.claude/mcp.json (Project root (project-scoped) or ~/.claude/settings.json (global))

{
  "servers": {
    "prep": {
      "command": "prep",
      "args": [
        "mcp"
      ]
    }
  }
}

Add "permissions": { "allow": ["mcp__prep"] } to settings.json to auto-approve all SourcePrep tools. Or run: claude mcp add prep -- prep mcp

Gemini CLI

~/.gemini/settings.json (Global config)

{
  "mcpServers": {
    "prep": {
      "command": "prep",
      "args": [
        "mcp"
      ],
      "trust": true
    }
  }
}

"trust": true auto-approves tool calls. Safe for SourcePrep (read-only tools).

OpenAI Codex

MCP config or AGENTS.md (Codex reads AGENTS.md natively)

{
  "mcpServers": {
    "prep": {
      "command": "prep",
      "args": [
        "mcp"
      ]
    }
  }
}

Codex also reads AGENTS.md — SourcePrep can auto-generate this file.