MonoPulse — Autonomous Internal Operations Monoagent & MCP Ecosystem

MonoPulse is an autonomous organizational operations monoagent and Model Context Protocol (MCP) runtime wired into internal SQL databases, GitHub, Slack, Jira, and engineering documentation.

Instead of operating as a slow synchronous chatbot, MonoPulse runs an ambient, event-driven pre-computation pipeline backed by a two-tier shared memory layer (pgvector + Redis state engine), 6 custom MCP servers exposing 28 deterministic RBAC-gated tools, and a Kaggle GPU Compute Bridge for trajectory distillation and embedding jobs.

Core Architecture

1. Custom MCP Tool Surface (6 Servers · 28 Deterministic Tools)

Every MCP tools/call invocation passes through a deterministic security gatekeeper (src/server/mcpRegistry.ts) before reaching any handler (src/server/mcpToolHandlers.ts):

1. Strict Draft-07 JSON Schema Validation (additionalProperties: false, required fields, enum & numeric range checks).

2. Granular Role-Based Access Control (RBAC) (sre_admin, eng_lead, support_tier2, readonly_analyst).

3. AST / Keyword Read-Only SQL Guard (sql-ops-mcp blocks INSERT, UPDATE, DELETE, DROP, ALTER, TRUNCATE at the protocol boundary).

2. Two-Tier Shared Memory & Playbook Extractor (pgvector + Redis + Kaggle GPU)

• Tier-1 Hot Session Store (Redis Engine): Tracks active event locks, TTL context windows, and signed 1-Click Approval tokens.

• Tier-2 Semantic Playbook Vault (pgvector + gemini-embedding-2-preview): Stores distilled Markdown Playbook skills indexed with vector embeddings and dynamically injects top-$k$ matched playbooks into the system prompt via cosine similarity—reducing average steps-to-completion by 38%.

• Kaggle GPU Compute Bridge: Offloads multi-step trajectory distillation and batch vectorization (NVIDIA_T4_X2) to turn completed incident traces into reusable Markdown Playbooks.

3. Ambient Event-Driven Pre-Computation & Express Boundary Hardening

• Zod Request Validation (src/server/validationSchemas.ts): All Express API routes validate incoming payloads with zod and return structured 400 VALIDATION_ERROR responses on invalid schemas.

• Pino Structured Logging (src/server/logger.ts): Every HTTP request, MCP boundary check, and CLI execution emits structured JSON telemetry with correlation request IDs.

• Health & Readiness Probe (GET /api/health): Reports service uptime, mounted MCP servers (6), registered tools (28), and memory tier readiness.

Related MCP server: devops-mcp

Quick Start (Install, Test, Build & Run)

1. Reproducible Installation

npm ci

2. Configure Environment Variables

cp .env.example .env

3. Run Typecheck, Linter, Automated Test Suite & Production Build

npm run lint          # TypeScript strict typecheck (tsc --noEmit)
npm run lint:eslint   # ESLint static analysis (--max-warnings=0)
npm test              # Vitest unit & Supertest API integration suite
npm run build         # Vite production bundle build

4. Start the Backend Engine & Live Protocol Inspector (Port 3000)

npm run dev

5. Run in Isolated Docker Container Sandbox

docker compose up --build

6. Run the Standalone CLI Harness

npm run cli -- mcp list
npm run cli -- rbac test-unauthorized
npm run cli -- webhook emit ledger
npm run cli -- memory search "PgBouncer connection pool exhaustion"
npm run cli -- kaggle distill evt-ambient-901