# Revvify

> Agentic AI marketing and sales. It plans, writes, schedules and publishes content
> across social and email, and works a sales pipeline from discovery through to
> attribution. A human approves anything that leaves the building.

This file exists because Revvify is built to be driven by an AI agent, not only by a
person. If you are an agent, everything below is the short version of what you can do
and what you cannot.

## Connect

- MCP server (Streamable HTTP): https://mcp.revvify.io/mcp
- Auth: OAuth 2.1 with PKCE. Dynamic client registration (RFC 7591) is supported, so
  no key is pasted by hand. Unauthenticated calls answer 401 with a
  `WWW-Authenticate` header naming the protected-resource metadata (RFC 9728).
- REST (for mobile and non-MCP clients): https://revvify.io/api/v1 — versioned,
  OpenAPI-described at /api/v1/openapi.json
- 37 MCP tools. Start with `get_capabilities`: it reports the plan, which channels
  are actually connected, which media kinds are available, and the consent table —
  so you can plan without guessing and without a failed call teaching you the limits.

## What an agent can do

Read brands, calendar, analytics, the lead inbox and the trend feed. Draft posts,
newsletters, replies and landing pages. Schedule and reschedule. Estimate the cost of
generating media before generating it. Ask for images and video. Read budgets and
operating mode.

## What no agent can do, ever

These are enforced server-side, not by prompt. An agent that tries gets a structured
error with a stable code and a remediation URL, not a refusal in prose.

- Send a DM, a cold email, or a reply to anyone. There is no tool for it — only
  `draft_lead_reply`, which leaves a draft for a human. Note the honest boundary:
  marketing email to opted-in subscribers DOES send on the schedule its campaign
  sets — the boundary is messaging a person, not email as such. Likewise,
  `schedule_post` and `publish_now` DO publish posts to channels the workspace has
  connected, because that is the product's purpose. Messaging a person is not.
- Follow or unfollow accounts. The tools do not exist.
- Persist generated video without a human accepting the preview first.
- Publish "controversy" content (STANCE or HUMOR spice modes) without human approval,
  regardless of autopilot settings.
- Touch another workspace. Every query is scoped by workspace and enforced in the
  database with row-level security, not in application code.

## How the AI is paid for

In this order, always: your own assistant over MCP → your own API key → prepaid
platform credits. Credits are a convenience and are never required by any feature.
Bring-your-own-key requests are billed to you by your provider at their cost; Revvify
does not mark them up and cannot see the key outside the worker that uses it.

## Pages worth reading

- https://revvify.io/mcp — the MCP reference: endpoint, all 37 tools, auth, errors
- https://revvify.io/for-ai-agents — why agent-native, and the contract behind it
- https://revvify.io/ — what the product does, for humans
- https://revvify.io/for-creators — the single-operator case
- https://revvify.io/for-agencies — the multi-client case
- https://revvify.io/legal/privacy — data handling
- https://revvify.io/legal/terms — terms

## Contact

hello@revvify.io