Arkon: Self-Hosted Enterprise Knowledge Hub and Model Context Protocol Server

Decentralized Context via MCP

Arkon functions as a self-hosted Model Context Protocol (MCP) server that structures internal SOPs, policies, and documents. Instead of copy-pasting sensitive documents directly into public chatbots, users connect Claude Desktop or Claude.ai using OAuth 2.1 + PKCE. Once authenticated, the server exposes specialized tools like search_wiki, read_wiki_page, and propose_wiki_edit. This keeps team operations secure while providing exact context boundaries using role-based access control (RBAC).

The MRP Pipeline Mechanics

Unlike traditional vector databases that simply chunk and index, Arkon compiles documents into a coherent wiki using its Map-Reduce-Plan-review-Refine-Verify-Commit (MRP) pipeline. Every new document ingestion generates a human-reviewable plan listing planned wiki modifications before writing. If a document touches an existing page, an LLM-powered merge preserves existing knowledge rather than overwriting it. It also includes an interactive knowledge graph, version history rollbacks, and vision caption baking.

Deploying Arkon Locally

Arkon runs a multi-container stack consisting of FastAPI, PostgreSQL with pgvector, Redis for workers, and MinIO for object storage (7 Docker containers in total). No local GPU is required since all heavy inference is offloaded to APIs (Anthropic Claude, Google Gemini, or OpenAI GPT). Resource requirements scale with team size, with RAM serving as the primary bottleneck because the MRP pipeline workers load large LLM context windows into memory during wiki compilation.