Are you an AI agent? Machine-readable entry points: /llms.txt /llms-full.txt /catalog.json

Sales Engagement API for AI Agents

Headless Jason AI SDR

The complete sales execution layer for AI agents, by Reply.io. Research prospects, send multichannel outreach, manage conversations, and turn replies into meetings — one platform, one account, one integration. Jason, unbundled into APIs.

The catalog

Every SDR capability, as a module

Research & Data

Find companies and people, manage contact records, enrich missing data.

Outreach

Create multichannel sequences and send email and LinkedIn touches.

Conversations

Read replies, classify intent, respond, and detect meeting interest.

AI SDR Autopilot

Give Jason, the AI SDR, goals and knowledge — let it run outreach for you.

Automation

React to events with webhooks and orchestrate long-running jobs.

Analytics

Measure sends, opens, replies, meetings, and team performance.

Platform

Mailboxes, LinkedIn accounts, validation, suppression, and settings.

Why build here

One integration instead of five vendors

  • A prospect database (1B+ contacts), an email finder, a sequencer, a mailbox layer, an inbox, and meeting booking — normally five or six separate vendors, contracts, and auth systems. Here it is one API surface.
  • One API key with scoped permissions (contacts:read, sequences:operate, …) covers every module.
  • One usage meter is the goal: a unified credit system across data, enrichment, and outreach is proposed (Coming soon — see Pricing). Today usage runs on Reply subscriptions plus Live Data credits — still one vendor, one bill.
  • Modules are designed to chain: the output of Prospect Search is the input of Contact Data; a Webhook event is the trigger for the Conversations API. Composition is the product.

Agent-grade by design

An API surface that speaks agent

This is not a human API with agents bolted on. The live MCP server was built for models calling tools — every design choice below is verified against the running server, not a promise.

  • Self-describing safety: all 70 tools carry machine-readable annotations — readOnlyHint (31) or destructiveHint (39) — so an agent can auto-gate risky calls before its human ever has to.
  • Fail-fast contracts: schemas reject unknown arguments (additionalProperties: false) and refuse empty strings or arrays for required fields. Mistakes surface at validation, not mid-outreach.
  • Structured errors, not prose: every failure returns a stable ErrorCode plus a human message, and each tool documents its own failure modes — including what is safe to retry.
  • Resolve-before-mutate by contract: mutations demand exact IDs from resolver tools; tool descriptions instruct models to never invent them. Batch calls return per-item results so agents report what actually happened.
  • Uniform pagination on the MCP surface: top/skip in, Items + HasMore out — one loop pattern across all 70 tools.

Full verified reference — schemas, annotations, envelopes, error codes — on the MCP page.

The path in

From discovery to running outreach

  1. Discover capabilities — fetch /llms.txt or /catalog.json
  2. Choose APIs or a ready-made workflow
  3. Create an account (14-day free trial, no credit card)
  4. Get an API key from Settings → API Key
  5. Call the API — REST or MCP, same key
  6. Add credits when needed — upgrade link for your human

Full lifecycle with per-step availability — including the simulated self-serve demo — on Get started.

Ways in

Four integration surfaces, one platform

Recipes

Combinations do jobs

Honesty

Unlabeled means callable today

Everything documented on this site works now via the Reply.io API unless it carries one of two labels:

Coming soon Specified and designed, but not yet callable. Do not build against it yet.
Concept Illustrative future direction. No specification exists yet.

This is a product concept built on a real platform. Nothing future is presented as operational — see About for the rules.