Technical Architecture

Overview

The Hive server is structured around a single wrapped tool catalog with multiple access surfaces layered on top.

Main Layers

Wrapped tool catalog

This is the live provider-backed tool surface used by the server runtime.

• 348 wrapped provider tools
• provider-backed handlers
• shared schema and metadata normalization

Root MCP endpoint

The root /mcp endpoint is the recommended AI-facing surface.

• compact discovery-oriented tools/list
• resource-based discovery through hive://tools, hive://categories, hive://providers, hive://toolsets, hive://task-canaries, hive://skills, and hive://status
• provider-tool execution through invoke_api_endpoint

Category-scoped MCP endpoints

These are narrower surfaces built from toolRegistry.ts.

• 10 scoped endpoints
• Hundreds of categorized tools
• direct tools/list results by category

REST API

The REST layer exposes:

• GET /api/v1/tools for the full live wrapped catalog
• POST /api/v1/execute for tool execution
• GET /api/v1/usage for plan, credit wallet, and key summary
• GET|POST|DELETE /api/v1/keys for dashboard key management
• GET /api/openapi.json and /docs for OpenAPI documentation
• /status, /api/health/*, /live, /ready, and /startup for health and canaries

Provider Routing

Requests are routed to provider-specific handlers based on tool metadata.

Current live provider inventory:

• CoinGecko
• Alchemy
• DeFiLlama
• GoPlus
• Codex
• CCXT
• Helius
• Tenderly
• Moralis

Discovery Model

There are two important discovery patterns:

1. Root discovery for task routing, schema lookup, and invoke_api_endpoint
2. Category-scoped discovery for smaller, direct tool lists

This distinction is the main reason the docs now separate Tools Reference from Live Catalog.

Related

• API Integration
• Tools Reference
• Data Sources