Introduction

OXYGEN Runtime API

OXYGEN exposes a narrow runtime surface for AI agents. Tool discovery, durable memory access, governed execution, and event streaming — reachable over HTTP or MCP, scoped to your organization.

PublicHTTP + SSEMCPOrg-scoped

One capability surface

GTM integrations and durable memory share a single schema, auth model, and execution contract. Agents learn it once.

Streamed by default

Every invocation returns an execution id. Subscribe to events for tool traces, checkpoints, and terminal state — no polling.

Bring your own planner

Claude Code, Codex, Cursor, or a custom agent keeps its reasoning loop. OXYGEN stays focused on governed GTM execution.

What you can build

The Runtime API is designed for long-running, tool-using agents. Anything you would hand to a GTM operator — sourcing leads, enriching accounts, drafting outreach, reading CRM state — can be invoked from your own code.

Custom agents over HTTP

Call POST /api/runtime/tools/{toolName} from any runtime. OXYGEN creates a session, executes the capability, and streams progress back.

IDE agents over MCP

Point Claude Code, Cursor, Codex, or Claude Desktop at https://oxygen-agent.com/api/mcp with a Bearer token and the same tools appear natively.

Where to go next

Quickstart

List capabilities, invoke a tool, stream the execution — all with curl.

Authentication

Create an org-scoped API key and pass it on every call.

Core concepts

How capabilities, sessions, executions, and events fit together.

API reference

Every endpoint, every parameter, every error body.

MCP

Use OXYGEN as a remote MCP server in Claude Code, Cursor, Codex, and Claude Desktop.

Errors

Status codes and the error envelope you'll see back.

Looking for OXYGEN's product docs?

These pages document the public API for external agents. For how to use OXYGEN inside the web app — sessions, plays, scheduled tasks, automation triggers — see the product site.

NextQuickstart