The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the Router MCP listing page.
Run the whole Intelliverse platform from any MCP-capable agent: create App IDs, mint tier API keys, ingest and search per-app knowledge bases, send email, generate media, check credits, and chat — without ever naming an LLM.
Verified 2026-07-12: initialize + tools/list over stdio, 15 tools.
Full setup guide with copy-paste configs: https://router.intelli-verse-x.ai/skills
| Tool | Auth | What it does |
|---|---|---|
chat | API key | Chat completion; model optional (Intelliverse default routes at your plan tier) |
list_models | API key | Tier aliases your key can call |
get_credits | API key | LLM daily budget + per-App-ID media credit wallets |
create_app / list_apps | management | App IDs — entry point for KB, wallets, per-app costs |
create_api_key / list_api_keys / revoke_api_key | management | Tier-gated key administration |
kb_ingest / kb_search / kb_chat / kb_documents | API key | App-scoped pgvector knowledge base |
send_email | API key | Per-App-ID email via the platform engine (Pro+, email.send10 SKU, hold → settle) |
generate_media / media_job_status | API key | Credit-metered image/video/voice/audio/3D jobs |
Env vars: ROUTER_API_KEY (Workspace → API Keys), ROUTER_MANAGEMENT_KEY
(Account → Management Keys; needed for app/key admin tools),
ROUTER_WORKSPACE_ID (default workspace for admin tools). Optional overrides:
EDGE_BASE_URL, KB_BASE_URL. Keys are free —
sign up, no card.
The JSON body is identical everywhere; only the file location / wrapper differs.
One-click: Add to Cursor
— or .cursor/mcp.json (project) / ~/.cursor/mcp.json (global):
Same JSON as Cursor, in claude_desktop_config.json
(macOS: ~/Library/Application Support/Claude/, Windows: %APPDATA%\Claude\).
.vscode/mcp.json:
Same JSON as Cursor, in ~/.codeium/windsurf/mcp_config.json.
Same JSON as Cursor, via the MCP Servers panel → Configure MCP Servers
(cline_mcp_settings.json).
~/.continue/config.yaml:
ChatGPT developer-mode apps connect only to remote MCP servers over HTTPS; this server is stdio-only today, so there is no ChatGPT path yet (hosted streamable-HTTP endpoint is on the roadmap). In the OpenAI ecosystem use the Codex CLI above.
skills/ ships SKILL.md playbooks that drive these tools end-to-end — copy them
into your agent's skills folder (~/.cursor/skills/ for Cursor,
~/.claude/skills/ for Claude Code):
| Skill | What it does |
|---|---|
intelliverse-router | Platform wiring — the 60-second create_app → key → KB loop |
nl-app-builder | Scaffold web/mobile apps wired to an App ID |
scaffold-game | Honest-scope Nakama game-backend stubs |
deploy-desktop | Electron/Tauri desktop apps delegating to IVX Agency |
conversational-avatar | Real-time voice + lip-sync avatars over LiveKit |
create_app — describe your app; get an App ID and recommended tier.create_api_key — mint a key bound to that app.kb_ingest — feed it docs or URLs.chat / kb_chat — start building. No model names, ever.send_email — tell your users it's live.server.json at the repo root is the payload published to the official
MCP Registry as
io.github.intelli-verse-x/router-mcp (MCPB bundle from the GitHub release).