Skip to main content
Reply.io is a sales-engagement platform an AI agent can operate end to end: find and manage contacts, launch multichannel outreach, triage and answer replies, pull performance data, and program Jason — Reply’s AI SDR — as a supervised autopilot. Everything runs behind one scoped API key across four access layers.

​
What an agent can do today

​
The four access layers

All four surfaces operate the same platform — same key, same objects, same limits. They differ in how an agent consumes them:
  • MCP — direct tool access for AI agents in MCP clients.
  • Skills — outbound expertise and guardrails as markdown packs; the skill decides what to do, the CLI or MCP does it.
  • CLI — terminal and automation interface with JSON output and safety flags.
  • API — low-level programmable integration; the exhaustive surface.
Not sure which fits? Choose your interface.

​
The human boundary today

An agent can operate everything behind the API key. Three things still need a human:
  1. Account creation — a human signs up (14-day free trial, no credit card required).
  2. API key issuance — created in the Reply app under Settings → API Keys and handed to the agent. Keys support granular scopes — see Authentication.
  3. Billing — plan upgrades and credit purchases happen in the Reply app.

​
Next steps

Choose your interface

MCP vs Skills vs CLI vs API — decision rules and a comparison table.

Agent safety rules

High-stakes gating, scoped keys, and rate-limit etiquette across all surfaces.

Drop-in system prompts

Ready-to-paste prompts for agents operating Reply.

Guides

Task-based guides for every outcome in the table above.