Setting up your own Claude to use these connectors? See Engineer Training — Claude Setup — it walks through adding every fleet connector (all except Xero) to your Claude, plus Claude Code + the branch workflow. This page is the concept reference; the setup steps live there.
Purpose: MCP (Model Context Protocol) is how we let AI assistants (Claude) safely talk to our tools — NinjaOne, Xero, Giacom, Wiki.js — through controlled "servers" that expose specific actions. This is bespoke Xzopia infrastructure, so it's part standard protocol, part our-own-build.
mcp.xzopiasecure.com) — all 11 connectors (NinjaOne, Xero, Giacom, Wiki.js, SentinelOne, TRMM, ITFlow, BookStack, Twenty CRM, M365Admin, …) on Node.js / PM2 / Nginx. S4 is now the live production box — untouchable without a backup + Simon.claudemcp.xzopiasecure.com) — the old MCP Live, drained to S4; no new work, wind-down pending. (History: S3 was the original live box; S4 was "dev-mcp" during build — that framing is now reversed.)github.com/Silentrunnerx/xzopia-deploy) is the source of truth for how our MCP servers are built and configured.docs/api-inventory.md in the repo — every API dependency mapped.mcp.xzopiasecure.com, Nginx): /ninjaone/mcp, /xero/mcp, /giacom/mcp, /wikijs/mcp, /sentinelone/mcp, /tacticalrmm/mcp, /itflow/mcp, /bookstack/mcp, /twentycrm/mcp, /m365admin/mcp. (S3's Caddy-fronted /ninjaone|/xero|/giacom is the old decommissioning endpoint.)/wikijs/mcp, mcp-servers/wikijs/src/index.js): wiki_search, wiki_get_page, wiki_list_pages, wiki_create_page, wiki_update_page (full-content replace), and wiki_patch_page(id, old_str, new_str[, expected_hash]) — surgical single-anchor edit: replaces one EXACTLY-once substring (0 or >1 matches → rejected, no write, returns the match count), optional SHA-256 expected_hash concurrency guard, re-fetches to verify the change landed. Use wiki_patch_page for small in-place edits (e.g. session-handoff id=15) — ~90% cheaper than a full-page rewrite and no clobber risk; use wiki_update_page only for page creation/restructure..env on the server — never hard-coded, never in the repo.127.0.0.1 only and are proxied via Nginx on 443; they're not directly reachable. Don't change this. (The old S3 box fronted the same pattern with Caddy — decommissioning, read-only reference now.)/backup-server --server server4 first, then verify all 11 connectors after (scripts/server4/mcp-healthcheck.sh or curl each /mcp). Server 3 is decommissioning — no changes except drain verification and wind-down..env from there; never paste them anywhere else.modelcontextprotocol.io for the concepts, then read our repo to see how a real server is structured..env, never the repodocs/api-inventory.md in the repoXzopia Limited (NI620366) | promote via Claude Code; do not hand-edit in the wiki.