The fragmentation of AI coding assistants has long been a productivity killer. You spend half your day re-prompting Claude about a decision you already made in Codex, or pasting context into ChatGPT that it should have inherited from your last session. Naetive claims to solve this by decoupling project understanding from the agent itself. Instead of maintaining separate memory states for each tool, Naetive creates a single source of truth that any MCP-compatible agent can access.

The Single Source of Truth

The core architecture rests on a .naetive folder stored directly in your repository. This folder contains the project's architectural decisions, build processes, and open questions. Because the understanding lives in the repo, it is not trapped inside a vendor's proprietary memory store. Every agentβ€”whether it is Claude Code, OpenAI's Codex, or standard ChatGPTβ€”connects to this same directory via the Model Context Protocol (MCP). There is no synchronization process because there are no separate states to sync; all agents read and write to the same underlying data structure.

Implementation Across Major Agents

Connecting the major players requires distinct setup steps but yields identical context availability. For Claude Code, users run a simple curl command (curl -fsSL https://naetive.ai/install | bash) on macOS, Linux, or WSL, which installs the MCP server and requires a one-time browser approval. Codex CLI users must manually edit ~/.codex/config.toml to add the Naetive server block with their specific key. ChatGPT and Claude on the web require a paid plan and Developer Mode (for ChatGPT) to add a custom MCP server URL. Crucially, for chat-based agents that lack direct file system access, Naetive provides a local tunnel program to bridge the gap between the cloud agent and the local project files.

Offline and Fresh Clone Support

A critical feature for enterprise or offline workflows is the ability to work without a live connection to the Naetive service. If you clone a repository on a new machine or use an agent that has never heard of Naetive, the .naetive folder remains accessible. Running naetive export injects pointers into standard agent configuration files like AGENTS.md or CLAUDE.md. This ensures that even without an active MCP server connection, the agent reads the committed project understanding on startup. It effectively turns the project's context into a portable, version-controlled asset rather than a transient session variable.

Key Takeaways

  • Naetive uses MCP to provide a unified context layer for Claude Code, Codex, ChatGPT, and Claude Desktop.
  • Project understanding is stored in a .naetive folder within the repo, enabling offline access and version control.
  • Setup varies by agent: CLI tools use config files or install scripts, while web agents require paid plans and custom MCP connectors.
  • The system eliminates the need to manually sync context between different AI coding assistants.

The Bottom Line

Naetive’s approach is a pragmatic fix for a messy reality. By treating project context as a version-controlled asset rather than a transient session state, it finally allows developers to switch agents without losing their train of thought.