Developer fatigue with fragmented documentation is hitting a breaking point, but a new open-source tool called Ohara aims to fix it by centralizing engineering guidelines. Created by ntrossat and detailed in a recent DEV.to post, Ohara allows teams to define architecture, design tokens, and coding standards in a single GitHub repository. Instead of manually updating CLAUDE.md or AGENTS.md files across dozens of projects, coding assistants like Claude and others fetch live instructions via the Model Context Protocol (MCP). This ensures that when an architect updates a guideline, every AI agent in every repository immediately follows the new rule.

The Problem: Documentation Drift and AI Inconsistency

The core issue Ohara addresses is the divergence of technical standards across multiple repositories. As the author notes, opening three repos at a typical company reveals three different architectures: one using a service layer, another embedding SQL in route handlers, and a third relying on a catch-all utils/ folder. When a team decides on a new standard, such as a unified API error format, that decision often gets lost in wiki pages or Slack announcements. Consequently, AI coding assistants, which rely on local context files, continue to generate code that violates the new standards because they never receive the updated instructions.

How It Works: Centralized Truth via MCP

Ohara operates by connecting your AI assistant to a central documentation server. Developers run claude mcp add --transport http ohara https://your-ohara/mcp to link the tool, followed by /ohara:init within a specific project. This process generates a .claude/rules/ohara.md file that contains pointers to the central guidelines rather than static copies. For example, the file directs the assistant to read guidelines/backend for FastAPI structure or design/ui-kit for component specs. When an assistant works on a task, it queries the Ohara server live, ensuring it always uses the most recently merged version of the documentation.

Closing the Loop: AI-Driven Doc Updates

Beyond reading guidelines, Ohara helps keep documentation synchronized with the code. After an AI assistant completes a feature, it can run /ohara:update to compare the latest code changes with existing docs. The assistant then proposes edits to project-specific documentation or broader architecture pages, bundling these suggestions into a pull request. A human reviewer must merge these changes, maintaining control while leveraging AI for maintenance. The system also flags stale pages; if a document hasn't been verified in 180 days or if related code changes, Ohara alerts the team, preventing the accumulation of outdated technical debt.

Key Takeaways

  • Ohara is an Apache-2.0 open-source tool deployable via a single Docker container.
  • It uses MCP to provide real-time access to centralized architecture and design guidelines.
  • AI assistants can propose documentation updates, which are reviewed and merged by humans.
  • The tool supports stale page detection to keep documentation accurate over time.

The Bottom Line

If your team is struggling to keep AI agents aligned with evolving standards, Ohara offers a practical, open-source solution to stop the copy-paste madness.

Get Started

You can try Ohara by running curl -fsSLO https://raw.githubusercontent.com/ntrossat/ohara/main/compose.yml && docker compose up -d. Once the container is up, navigate to http://localhost:8000 to set up the GitHub App. The project is currently available at github.com/ntrossat/ohara, with a live demo of its own docs hosted at ohara.trossat.com/docs.