This project provides a local, open-source server that indexes documentation from various sources to create an up-to-date, version-aware knowledge base for AI coding assistants. It aims to combat AI hallucinations and stale knowledge by grounding AI responses in accurate, context-specific documentation, thereby boosting developer productivity.

How It Works

The server scrapes documentation from websites, GitHub, package managers (npm, PyPI), and local files. It employs semantic chunking to preserve logical document structure and then generates embeddings using a configurable selection of LLM providers (OpenAI, Google, Azure, AWS). These embeddings, combined with vector similarity and full-text search, enable precise, version-aware retrieval of information for AI agents via the Model Context Protocol (MCP).

Quick Start & Requirements

• Recommended: docker compose up -d after cloning the repo and setting OPENAI_API_KEY in .env.
• Alternative: npx @arabold/docs-mcp-server@latest (data persistence requires Docker).
• Prerequisites: Docker Desktop (recommended), OpenAI API key (or other supported provider credentials).
• Setup: Minimal setup time, primarily configuring API keys.
• Links: GitHub Repo, Web Interface, MCP Client Configuration.

Highlighted Details

• Supports multiple embedding providers including OpenAI, Google Gemini/Vertex AI, AWS Bedrock, and Azure OpenAI.
• Features semantic chunking to maintain context and structure of documentation.
• Offers a user-friendly web interface for managing and searching indexed documentation.
• Designed for local execution, ensuring data privacy.

Maintenance & Community

The project is community-driven and open-source. Further details on development and contribution can be found in ARCHITECTURE.md. Issues can be opened for questions or suggestions.

Licensing & Compatibility

Licensed under the MIT License, permitting commercial use and integration with closed-source projects.

Limitations & Caveats

The npx installation method does not persist data between runs. Local file indexing requires careful path management, especially when using Docker, necessitating volume mounts.