DSOM-MCP-ARCHITECTURE.md - Model Context Protocol Native Server

Research and architectural blueprint for converting the DSOM Palace into a native MCP server for AI clients. Deep State of Mind framework.

“Context is not a static text block; it is an interactive resource. Let the AI query the Palace itself.”

---

1. The Context7 Inspiration

Traditional DSOM interactions required passing context into the chat explicitly via .dsom state files, palace_update_proposals, or the human manually injecting references. Services like Context7 demonstrated a better approach: RAG (Retrieval-Augmented Generation) exposed natively to the AI client.

Instead of pushing context to the AI, the AI pulls exactly what it needs from the Sovereign Markdown Palace through a standardised API: the Model Context Protocol (MCP).

2. Core Architecture of DSOM-MCP

We are building a native Python MCP Server (dsom-mcp-server) that completely replaces the need for external third-party syncs for local development.

2.1 The Transport Layer (STDIO)

The MCP server operates as a local subprocess spawned by your AI editor (Cursor, Google Jules, Claude Desktop). Communication happens over stdio using JSON-RPC.

• Rule: The MCP script must strictly output JSON-RPC to stdout. All logs or debugging MUST be written to stderr. Using print() without redirecting to sys.stderr will instantly break the protocol.

2.2 Framework & Ecosystem (Rule 16 Compliance)

• Executor: The server is executed exclusively via uv run to maintain Python environmental isolation.
• Framework: We utilize the fastmcp (or official mcp) Python SDK to define resources and tools asynchronously.

3. The 3 Pillars of the DSOM-MCP Server

Pillar A: Exposed Resources (Memory & State)

Resources are static or dynamic data blobs the AI can “read” at will without using a tool.

• dsom://state/current → Serves .agents/brain/current_state.dsom
• dsom://state/task → Serves .agents/brain/task.md
• dsom://state/walkthrough → Serves .agents/brain/walkthrough.md

When an AI boots up, it reads these URIs immediately to achieve the “Genesis Read” without the human typing a single prompt.

Pillar B: Exposed Tools (The Rituals)

Tools are executable functions. We expose our existing DSOM automation to the AI natively.

• search_palace(query): Executes a semantic or grep search across docs/.
• palace_sync(): Triggers the EOD spatial reflection engine natively.

Pillar C: Execution Bridge

Because Windows (T1) and WSL2 (T2) possess execution boundaries, the MCP server must detect its environment and invoke the tools/ Bash or PowerShell scripts appropriately.

4. MCP Client Configuration Example

To attach the DSOM-MCP server to Claude Desktop (or Cursor), the human operator modifies their client config (claude_desktop_config.json):

{
  "mcpServers": {
    "dsom-palace": {
      "command": "uv",
      "args": [
        "run",
        "--with", "mcp",
        "tools/mcp/server.py"
      ],
      "env": {
        "DSOM_ROOT": "/absolute/path/to/project"
      }
    }
  }
}

5. Security Posture

• Zero-Network Surface: The server runs exclusively on local stdio. No HTTP ports are opened.
• No External Exfiltration: Unlike passing codebase context to third-party RAG providers, all semantic searching and reading happens strictly on the local machine.

---

Standard: DSOM For My AI Protocol v6.1 | Harisfazillah Jamel | LinuxMalaysia

---

Deep State of Mind (DSOM) For My AI Protocol | Harisfazillah Jamel (LinuxMalaysia) | 2026-08-14 Standard: UK English | DBP-standard Bahasa Melayu Malaysia (Piawai) | GNU General Public License v3.0