KodHau

Product Introduction

1. Definition: KodHau is a specialized Model Context Protocol (MCP) server designed to bridge the gap between static code analysis and institutional memory. It functions as a context-injection layer for AI coding agents, extracting "tribal knowledge" from historical version control data—including pull request (PR) history, architectural design decisions, and peer review comments—to provide LLMs with the underlying rationale behind existing code structures.

2. Core Value Proposition: KodHau exists to solve the "context deficit" inherent in modern AI development tools. While standard AI agents like Cursor or Claude Code can parse current files, they lack awareness of why specific constraints were implemented or why certain approaches were previously rejected. By indexing a team's undocumented decision-making history, KodHau prevents AI-generated regressions and ensures that agentic workflows respect complex architectural boundaries and legacy constraints.

Main Features

1. Automated PR History & Tribal Knowledge Extraction: KodHau utilizes a deep-scanning engine that parses closed pull requests to identify the "why" behind the code. It specifically targets architecture decisions, rejected implementation paths, and nuanced feedback found within code review comments. This process transforms a repository’s historical commit data into a structured knowledge base of latent design patterns and team-specific constraints that are typically never documented in official README files or wikis.

2. The kodhau_get_wisdom MCP Tool: This is the primary interface between the AI agent and the tribal knowledge repository. When integrated via the Model Context Protocol, the kodhau_get_wisdom function allows an AI agent to proactively query for context regarding a specific file or function before generating code. It injects relevant historical data—such as "this API contract was intentionally kept lazy for performance reasons"—directly into the agent's prompt context window, ensuring the output aligns with established engineering standards.

3. Local-First Privacy Architecture: Security is maintained through a local execution model where the MCP server runs on the developer's machine. KodHau ensures that sensitive source code and proprietary PR data never leave the local environment or the user's controlled infrastructure. This architecture makes it compatible with enterprise security requirements while remaining accessible to solo developers through any client that supports the Model Context Protocol (MCP).

Problems Solved

1. AI-Driven Regressions and "Blind Refactoring": A common pain point occurs when an AI agent suggests "cleaning up" code that appears redundant but is actually serving a critical, undocumented purpose (e.g., handling a specific edge case or maintaining backward compatibility). KodHau stops agents from recommending destructive changes by providing the historical context that explains the necessity of the current implementation.

2. Target Audience:

• Senior Software Engineers & Tech Leads: Who need to ensure AI tools don't violate long-term architectural integrity or repeat past mistakes.
• Solo Developers: Who are working on complex open-source projects or legacy codebases where they need to understand the original author's intent.
• Engineering Teams: Organizations looking to scale institutional knowledge without requiring exhaustive manual documentation for every minor design decision.
• Enterprise Organizations: Companies that utilize AI agents but require strict adherence to internal compliance and documented historical constraints.

3. Use Cases:

• Legacy Code Refactoring: When an agent needs to update old modules without breaking hidden dependencies.
• Onboarding to New Repositories: Helping a developer (and their AI assistant) quickly understand the "unwritten rules" of a large codebase.
• Complex Bug Squashing: Providing the agent with previous PRs that attempted to fix similar issues, highlighting what worked and what failed.
• API Maintenance: Ensuring that proposed changes to public or internal interfaces do not violate historical contract decisions.

Unique Advantages

1. Differentiation from Standard RAG: Traditional Retrieval-Augmented Generation (RAG) tools focus on the current state of the codebase (the "what"). KodHau focuses on the evolution of the codebase (the "why"). By prioritizing PR comments and rejected approaches, it provides a layer of intelligence that standard vector search over current files cannot replicate.

2. Key Innovation: Historical Intent Injection: The specific innovation lies in treating the PR history as a primary data source for AI context. While most tools look at the code as it exists today, KodHau treats every comment and commit message as a constraint or an instruction. This "historical intent" ensures that the AI agent acts as an informed member of the team rather than an outside observer guessing at the logic.

Frequently Asked Questions (FAQ)

1. What is an MCP server, and how does it work with KodHau? The Model Context Protocol (MCP) is an open standard that enables AI agents to connect to external data sources. KodHau operates as an MCP server, meaning it acts as a standardized "plugin" for tools like Cursor and Claude Code. Once connected, it gives the AI agent a specific tool (kodhau_get_wisdom) to fetch relevant historical context automatically during the coding process.

2. Does KodHau require my code to be uploaded to their servers? No. KodHau is built with a local-first philosophy. The analysis of your PR history and the execution of the MCP server happen on your local machine. Your code and internal team discussions remain private, ensuring compliance with strict data security and intellectual property policies.

3. How does KodHau improve AI accuracy compared to just using Cursor's built-in indexing? While Cursor indexes the current file structure and code symbols, it often lacks the context of "why" a line of code was written a certain way. KodHau supplements Cursor's index with data from closed PRs and review comments. For example, if a line of code looks inefficient but was written that way to solve a specific race condition documented in a PR from two years ago, KodHau will inform the AI, whereas standard indexing might suggest "optimizing" it and reintroducing the bug.

4. Which AI clients are compatible with KodHau? KodHau is compatible with any development tool or IDE that supports the Model Context Protocol. This includes popular AI-native editors like Cursor, terminal-based agents like Claude Code, and other emerging MCP-compliant clients.