Skip to main content

Olympus MCP Server

The Olympus MCP (Model Context Protocol) Server is the primary governance and automation interface for AI coding agents (Claude Code, Gemini Code Assist). It transforms AI agents from "text generators" into "autonomous engineers" by providing safe, verified access to the Olympus Cloud infrastructure.

Core Capabilities​

🛡️ The Guardian Module (Anti-Hallucination)​

The MCP server implements a Zero-Trust execution model. An AI cannot simply state that a task is finished; it must provide a proof_command (e.g., cargo test) that the MCP server executes locally.

  • State Machine: Tracks Acceptance Criteria (A/C) status in a persistent local JSON store.
  • Empirical Proof: Only successful tool execution "unlocks" A/C items.
  • Architectural Peer Review: Bridges the CLI to LangGraph Roman God agents (Jupiter, Minerva) for mandatory sign-offs before PR submission.

🔍 Exhaustive RAG & Discovery​

Provides high-efficiency semantic search over the entire monorepo and documentation.

  • search_code / search_docs: Powered by Cloudflare Vectorize and Workers AI (Tier 1 - Free).
  • rag_index_update: Allows agents to autonomously rebuild semantic indexes after large features are completed.

🛠️ Structural Refactoring​

Beyond simple string replacement, the MCP provides AST-aware tools to ensure code integrity.

  • refactor_rename_symbol: Precisely renames variables, functions, and classes across files using boundary-aware logic.

📈 FinOps & Reliability​

  • Session Circuit Breaker: Automatically pauses an autonomous agent if it enters an infinite failure loop (e.g., more than 5 consecutive test failures) to prevent API budget drain.
  • Token Optimization: Summarizes large log outputs using cheap local models before sending them to expensive Tier 5/6 models.

Tool Inventory (101 Tools)​

The server exposes 101 tools across 21 categories, including:

  • Governance: Guardian state machine, standards audit, peer review.
  • Operations: Cloud Build status, canary promotion, secret provisioning.
  • Diagnostics: Full-stack log correlation, Flutter heuristic analysis.
  • Environment: IoT hardware simulation, Edge network partition simulation.

AI Configuration​

Configure the server in your AI CLI settings:

{
  "mcpServers": {
    "olympus": {
      "command": "node",
      "args": ["dist/index.js"],
      "env": {
        "OLYMPUS_WORKSPACE": "${workspaceFolder}",
        "MCP_LOG_LEVEL": "info"
      }
    }
  }
}