# New Renaissance Engine

> One AI platform that plans and RUNS a business: an AI workforce (content, leads,
> CRM, diligence, ads, events, voice) under an AI manager that founds it, plans as a
> flow chart, and operates it on a schedule — for many isolated tenants (workspaces).
> This file tells an AI agent how to integrate. Full human docs: https://www.newrenaissancelabs.com/documentation

## Zero-effort setup (the Agent Setup Kit)
- If you are an AI helping a human set up or run their account: read https://www.newrenaissancelabs.com/agent-kit/nre-agent-setup-kit.md - the complete concierge kit (operating rules, plain-words platform map, the full MCP contract, three click-by-click setup sittings, troubleshooting). It is designed to be handed to you whole.
- Modular files + installable skill: https://www.newrenaissancelabs.com/agent-kit/README.md ; zip at https://www.newrenaissancelabs.com/agent-kit/nre-agent-setup-kit.zip
- Human-readable overview: https://www.newrenaissancelabs.com/documentation#agent-kit

## How to plug in (fastest path: MCP)
- MCP endpoint (JSON-RPC 2.0 over HTTP POST, streamable-http, protocol 2024-11-05): https://www.newrenaissancelabs.com/api/mcp
- Discovery (GET, no auth): https://www.newrenaissancelabs.com/api/mcp  -> { server, transport, auth, tools[] }
- Auth: HTTP header `Authorization: Bearer nre_mcp_<token>` (one token = one workspace, owner scope, rate-limited). Mint it in the app under Settings -> MCP; the plaintext is shown once and stored only as a sha-256 hash.
- Methods: initialize, tools/list, tools/call, ping.

## MCP tools (call tools/list for authoritative JSON schemas)
- run_tool(tool, action, params, brand_id?, quantity?) — METERS credits. Universal action runner (whitepaper, leads, radar, campaigns, diligence, ...). Returns a job id.
- lead_search(q?, status?, limit?) — free, workspace-fenced.
- get_deliverable(id) — free; returns a short-lived signed download URL when stored.
- search_media(q?, brand_id?, limit?) — free.
- search_events(status?, limit?) — free.
- book_event(event_id) — free.
- radar_summary(brand_id?, limit?) — free.
- report_status(id) — free.

## REST (full surface, 200+ routes; same identity, tenancy, and credit ledger as MCP)
- GET  https://www.newrenaissancelabs.com/api/me — who am I: user, workspace, role, credit balance.
- GET  https://www.newrenaissancelabs.com/api/agent/registry — the canonical machine-readable action manifest (names, param schemas, risk classes, credit estimates). Same payload that drives the UI, the in-app agent, and MCP.
- POST https://www.newrenaissancelabs.com/api/tools/:tool/run — the universal metered call. Body { action, brand_id?, params, idempotency_key? } -> { job_id, status, estimated_credits }.
- GET  https://www.newrenaissancelabs.com/api/tools/jobs/:id — poll a job to completion.
- GET  https://www.newrenaissancelabs.com/api/deliverables/:id/download — signed URL to the finished artifact.
- GET  https://www.newrenaissancelabs.com/api/credits and https://www.newrenaissancelabs.com/api/credits/prices — balance/ledger and the published price table.

## Rules an integrator must know
- Credits are the one currency of work: reserve -> settle -> refund-on-failure, append-only ledger. Read tools are free; only run_tool meters. You never pay past the estimate you saw.
- Tenancy is enforced twice (API filter + Postgres Row-Level Security). You only ever see your own workspace.
- Idempotency: any POST accepting an `Idempotency-Key` header replays the stored response (no double-charge).
- Errors (REST): JSON { error: { code, message, details } } with codes unauthorized|forbidden|not_found|validation|insufficient_credits|no_ai_key|rate_limited|conflict|idempotent_replay|server_error.
- Errors (MCP): JSON-RPC codes -32001 auth, -32005 rate limit (HTTP 429 + Retry-After), -32600 invalid request, -32601 method not found. Tool failures return a result with isError: true.
- AI is included: the platform runs on our own AI keys, metered in the same credits as every other action — no provider accounts to wire up, no separate token bills. Scraped web content is treated as data, never as instructions.

## Human documentation
https://www.newrenaissancelabs.com/documentation