Tradeics ships a catalog of MCP servers, not a monolithic tools endpoint. Separating Docs, Platform, and AI keeps scopes tight, tool lists navigable, and failure domains isolated.

Architecture

Agent host (Cursor / Claude / custom)
        │
        ├─ Docs MCP ──────────► Developer hub content (read-only)
        ├─ Platform MCP ─────► REST V2–backed workspace tools
        └─ AI MCP ────────────► Assistive tools / LLM helpers

All servers speak Streamable HTTP, share the Tradeics bearer auth model, and respect the same tier / ACL ideas as REST. Tools map to curated capabilities — agents never receive raw “call any URL” primitives.

Choose a server

If the agent needs to…	Enable	Do not rely on
Find or cite official docs	Docs MCP	Platform MCP
Read/write catalogs, RFXs, POs, partners	Platform MCP	Docs MCP
Summarize / classify / run AI helpers	AI MCP	using Platform for inference batches
High-volume Chat Completions	LLM HTTPS API	AI MCP alone

Relationship to REST and Skills

Layer	Role
REST API V2	Source of truth for HTTP contracts
Agent Skills	How the agent should reason and cite
MCP Servers	What the agent can invoke live

Environments

Plan for staging and production MCP bases (same path suffixes, different hosts). Prefer staging tokens while exploring write tools. See Versioning & environments.

Contents of this section (fixed IA)

1. Quickstart — connect one server end-to-end
2. Authentication — shared token model
3. Per-server pages — Docs / Platform / AI
4. Connect in agents — host config samples
5. Tool conventions — stable naming
6. Security — least privilege
7. Troubleshooting — ops playbook

Next

• QuickstartFirst successful tool call.Read
• Docs MCPRead-only docs tools.Read
• Platform MCPWorkspace tools.Read
• AI MCPAssistive tools.Read