Quickstart

See docs/quickstart.md.

Architecture

See docs/architecture.md.

agentic-platform-mcp

A production-grade agentic AI platform built on the Model Context Protocol (MCP):

• Multi-server MCP host (SharePoint / ServiceNow / KB + more later)
• Full MCP capabilities: Tools, Resources, Prompts (and later Sampling/Elicitation where needed)
• Local open-source LLM inference (vLLM or alternative) with tool/function calling
• LangGraph orchestration layer
• Retrieval routing layer (dual-encoder + cross-encoder rerank)
• Kubernetes deployment on K3s (DGX Spark)
• Observability: OpenTelemetry traces/metrics/logs + dashboards

Why this exists

Most “agent demos” break when you add:

• multiple tool providers,
• policy controls (PII, permissions),
• scaling/deployment,
• and monitoring.

This repo is designed as a staff-level system: modular, secure-by-default, and deployable.

High-level architecture (MCP-centric)

User → Host API/UI → Agent Orchestrator (LangGraph) → MCP Clients (1 per server) → MCP Servers
The LLM is the reasoning engine; the agent is the control loop + state manager.

Scope (initial)

MCP Servers (phase 1)

• mcp-kb: internal knowledge base (policies, runbooks) via resources + search tool
• mcp-sharepoint: document retrieval + metadata resources + search tool
• mcp-servicenow: incident/query tools + read-only resources for tickets

Host + Agent (phase 1)

• MultiServer MCP host with dynamic discovery
• Tool approval + policy gating (PII logging checks)
• Conversation loop (CLI first; API later)

Quickstart (will be filled as we build)

• Local dev: Python + uv
• Run servers (SSE) and host (CLI)
• Deploy to K3s

Repo map

• src/host/ — multi-MCP host + agent loop + LLM adapter
• src/servers/ — MCP servers (KB/SharePoint/ServiceNow)
• src/common/ — shared auth, schemas, policies, utilities
• deploy/ — Docker + K3s manifests/helm
• observability/ — OpenTelemetry + dashboards
• docs/ — architecture, decisions, roadmap

Roadmap

See docs/roadmap-14-days.md.

License

Apache-2.0 (recommended for industry-friendly reuse).# agentic-platform-mcp A production-grade, open-source agentic AI platform built on Model Context Protocol (MCP). Supports multi-MCP servers, local LLM inference, agent orchestration (LangGraph), retrieval-augmented generation, Kubernetes deployment, and full observability.