ISA MCP Platform

Enterprise-grade Model Context Protocol (MCP) platform for tool registration, discovery, invocation, and external MCP aggregation.

Overview

ISA MCP provides:

• Unified tool/prompt/resource registry
• Skill-based semantic discovery
• Graceful semantic-to-lexical search fallback
• Human-in-the-loop (HIL) workflows
• External MCP server aggregation
• Standalone enterprise MCP servers
• Multi-tenant isolation and security controls

Latest Updates (2026-05-09)

• Kubernetes liveness endpoint added at /live so probes can distinguish process liveness from dependency readiness.
• Runtime branding support added for white-label deployments and non-interactive rebrand flows.
• Backup/restore scripts and SOPs added for MCP runtime recovery.
• Disaster-recovery docs started with an operator dependency map and chart-hardening notes.
• Health semantics refresh so /health now distinguishes healthy, degraded, and unready states.
• Search degradation support so the platform falls back to lexical search if the semantic model dependency is unavailable.
• Enterprise MCP server docs for the dedicated API gateway, database, filesystem, and message queue servers.
• Security and reliability docs refresh covering response headers, rate limiting, and degraded-mode behavior.

Platform Capabilities

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                         ISA MCP PLATFORM                        │
├─────────────────────────────────────────────────────────────────┤
│  Tools / Prompts / Resources                                   │
│  ↓                                                             │
│  Services: Tool, Prompt, Resource, Skill, Search, HIL, Sync   │
│  ↓                                                             │
│  Aggregator: connect and route to external MCP servers         │
└─────────────────────────────────────────────────────────────────┘

Core Features

Skill-Based Search

Natural-language query routing through skill classification and ranked matches. Semantic search is the default path; if the model dependency is unavailable, MCP degrades to lexical search and reports that mode through /health and search metadata.

results = await search("schedule a meeting with John")

Human-in-the-Loop

Supports:

• authorization
• input collection
• review
• combined input + authorization

Aggregator Service

Register, connect, monitor, and route to external MCP servers with namespace-aware tool routing.

Standalone Enterprise Servers

The repo also ships smaller MCP servers for focused infrastructure domains:

• api_gateway_mcp
• database_mcp
• filesystem_mcp
• message_queue_mcp

Each exposes the standard MCP surfaces:

• GET /health
• GET /.well-known/mcp-server-card.json
• POST /mcp

Marketplace

Install, manage, and discover tool packages with org-scoped isolation:

# Search packages
curl -X POST http://localhost:8081/api/v1/marketplace/search \
  -H "Content-Type: application/json" \
  -d '{"query": "data analysis", "category": "analytics"}'
 
# Install a package
curl -X POST http://localhost:8081/api/v1/marketplace/install \
  -H "Content-Type: application/json" \
  -d '{"name": "isa-analytics-tools", "version": "1.2.0", "auto_connect": true}'
 
# Batch install
curl -X POST http://localhost:8081/api/v1/marketplace/install/batch \
  -H "Content-Type: application/json" \
  -d '[{"name": "isa-analytics-tools"}, {"name": "isa-browser-tools"}]'
 
# List installed
curl http://localhost:8081/api/v1/marketplace/installed
 
# Check for updates
curl http://localhost:8081/api/v1/marketplace/updates

Package lifecycle: install → enable/disable → update → uninstall. All operations are scoped per organization.

Production Security

Enterprise-grade security hardening across the platform:

• Authentication — Unified auth middleware supporting JWT Bearer tokens and API keys (X-API-Key, X-MCP-API-Key). API keys in query parameters are explicitly rejected.
• Multi-org context — X-Organization-Id header for org switching with validation against authorized orgs.
• Command safety — Allowlist-based command validation with blocked patterns for injection, code execution, dangerous file ops, and sensitive file access. Max command length: 8192 bytes.
• Input validation — Query length limits (≤1000 chars), result limits (≤50), threshold bounds ([0,1]), and regex-validated operation IDs for SSE streams.
• SSE injection prevention — Operation IDs validated against ^[a-zA-Z0-9_-]{1,128}$ to prevent path traversal.

Security & Operations

• Security - auth, authorization levels, SSRF/traversal/injection controls
• Multi-Tenant - org scoping, global vs org-local resources
• Reliability - health semantics, retries, degraded modes, and overflow protection

API Endpoints

Base URL: http://localhost:8081

Documentation

Getting Started

• Quick Start
• Configuration

Components

• Tools
• Prompts
• Resources
• Skills

Advanced

• Search & Discovery
• Human-in-the-Loop
• Progress Tracking
• Server Aggregation