Arkon - The Open-Source Enterprise AI Knowledge Hub & MCP Server

Arkon is a self-hosted, enterprise-grade knowledge management layer that bridges organizational data and AI clients. It runs as a centralized MCP Server (Model Context Protocol), compiling your SOPs, policies, and internal docs into a structured, traceable knowledge wiki - then serving that wiki to Claude and other LLMs through a single permission-scoped endpoint.

🚀 Why Arkon?

In most organizations, AI adoption is fragmented. Employees copy-paste documents into chatbots, producing inconsistent context, security risks, and duplicated work.

Arkon treats AI as a managed organizational resource. Every employee gets the right context, automatically and securely - filtered by their department and role.

---

✨ Key Features

🧠 Intelligent Knowledge Wiki - the MRP Pipeline

Unlike a vector database that just chunks and indexes, Arkon's MRP pipeline (Map → Reduce → Plan-review → Refine → Verify → Commit) actually compiles documents into a coherent wiki of interlinked pages.

• Plan review before write: every ingestion produces a human-reviewable plan listing which wiki pages will be created or updated. Editors can regenerate the plan with feedback before any page is written.
• Page merge instead of overwrite: when a new source touches an existing wiki page, content is LLM-merged so prior knowledge is never lost.
• Traceable claims: every page records the source documents it was compiled from.
• Image-aware: vision captions are baked into source text before compilation, so wiki pages reference the right images in the right places.
• Resumable: drafts persist mid-pipeline; a crashed run resumes without re-doing the expensive LLM work.

📚 Wiki Browser & Knowledge Graph

• Three-panel layout: page tree, content, backlinks & outlinks.
• Full-text + semantic (pgvector) search.
• Interactive knowledge graph visualization (per-scope or global).
• Wikilink-style cross-references between pages.
• Version history and rollback on every page.
• Draft proposal → editor review → approval workflow.

🏢 Department & Global Scopes

Precise context boundaries with their own scoped wiki, document set, and department rosters.

• Department-level isolation for HR, Legal, Engineering, etc.
• Global scope for company-wide SOPs, policies, and general knowledge.
• Hard scope enforcement: members only see knowledge from their assigned departments or the global realm - at the API, MCP, and search layers.

🛂 Fine-Grained RBAC

Role-based access control at the department and global levels.

• Built-in roles: Viewer · Contributor · Editor · Admin to control draft, edit, and setting access.
• Granular permissions (doc:read:own_dept, wiki:edit:all, org:settings:manage, ...).
• Audit log for every privileged action - settings changes, plan approvals, and role updates.

🔌 MCP Server for Claude & Other AI Clients

Employees connect Claude Desktop or Claude.ai to Arkon via OAuth 2.1 + PKCE — just add the server URL and sign in through the browser. No manual token copying required. The MCP server exposes:

• Wiki tools - search_wiki, read_wiki_page, list_wiki_pages, read_wiki_index.
• Source drill-down - get_source, get_source_outline, get_source_pages, list_sources.
• Edit workflow - propose_wiki_edit, edit_wiki_page, list_pending_drafts, review_draft, approve_draft, reject_draft.
• Discovery - list_knowledge_types, get_knowledge_type_docs.

All tools enforce per-token scope (department, knowledge type, source list).

🧰 AI Skills Distribution

Upload custom agent packages once and distribute them across the org.

• Versioned skill packages (.zip with SKILL.md).
• Department-scoped visibility.
• Contribution workflow for end-user updates.

🤖 Pluggable AI Providers

Catalog-driven selection - admins pick from a curated list with context window, cost, and capability metadata on display.

• LLM: Anthropic Claude (Opus / Sonnet / Haiku 4.x), Google Gemini (3.x Pro / Flash / Flash-Lite, 2.5), OpenAI (GPT-5.x, GPT-4.x).
• Embedding: Google gemini-embedding-*, OpenAI text-embedding-3-* - switchable with online re-embed migration (active model atomically flipped on completion, no zero-result search window).
• Vision: Google Gemini Flash, OpenAI GPT-4o family.

🔒 Privacy & Security First

• Self-hosted. Deploy on-prem or in your private cloud via Docker.
• No telemetry. Outbound traffic goes only to the AI provider you choose.
• Encrypted at rest. API keys stored with Fernet encryption in PostgreSQL.

---

🛠️ Tech Stack

• Backend: FastAPI · PostgreSQL + pgvector · Redis (arq workers) · MinIO
• Frontend: Next.js · Tailwind CSS
• AI integration: Model Context Protocol via FastMCP
• Document parsing: PDF, DOCX, DOC, plain text, URLs, embedded images

---

💻 Server Requirements

Arkon runs 7 Docker containers (PostgreSQL + pgvector, Redis, MinIO, FastAPI API, 2 ARQ workers, Next.js frontend). The table below provides recommended configurations based on team size:

	Starter	Team	Enterprise
Team size	1 – 20	20 – 100	100+
vCPU	2 cores	4 cores	8+ cores
RAM	4 GB	8 GB	16+ GB
Storage	40 GB SSD	100 GB SSD	250+ GB NVMe SSD
OS	Ubuntu 22.04+ / Debian 12+	Ubuntu 22.04+ / Debian 12+	Ubuntu 22.04+ / Debian 12+
Use case	Evaluation / small teams	Departmental deployment	Organization-wide rollout

[!NOTE]

• RAM is the primary bottleneck — the MRP pipeline workers load large LLM context windows into memory during wiki compilation.
• Storage scales with your document corpus — pgvector indexes, MinIO file storage, and PostgreSQL WAL logs are the main consumers.
• All AI inference happens externally (Anthropic