Overview

Gandalf is the local control console and standardizer for AI agent environments (Claude Code, OpenAI Codex).

Just as Docker solved “it works on my machine” for runtime containers, Gandalf solves “it works on my agent” for AI-assisted engineering teams.

Why Gandalf?

Modern software teams develop primarily using AI agents. Previously, onboarding a developer meant installing an IDE and a few extensions. Today, team members need:

  1. Shared MCP Servers (staging database connectors, internal API search, log query tools)
  2. Standardized Skills (team coding guidelines, PR reviewer skills, prompt recipes)
  3. Guardrail Hooks (lint before save, sensitive file protection)

Without a unified declarative setup, each developer’s agent has different capabilities, missing MCP tools, or conflicting configurations.

Gandalf provides Agent Environment as Code: declare everything in a single gandalf.toml in your repository root, verify drift with gandalf check, and synchronize local setups with gandalf apply.

Key Capabilities

  • Agent Environment as Code: Declare MCP servers, skills, and hooks in a Git-tracked gandalf.toml.
  • Non-Destructive Smart Merge: Safely injects team tools without deleting or modifying personal settings.
  • Drift Detection & CI: Detect missing agent capabilities automatically locally or in CI pipelines.
  • Instant Safety Rollback: Pre-apply backups allow one-click rollback with gandalf restore.