Skip to content

Use EchoRelay from an AI agent

EchoRelay is built to be driven by AI agents, not just humans. An agent can create and manage your projects, lines, endpoints and API keys, send test traffic, and inspect delivery — all through the same surfaces you'd use by hand.

There are two equivalent ways in; most agents use the first, and fall back to the second.

We publish a public Agent Skill — a portable SKILL.md package that teaches any capable agent (Claude, Claude Code, or anything that reads Agent Skills) how to integrate EchoRelay end-to-end, MCP-first with a REST fallback.

The skill covers the core recipes: your first endpoint (project → line → endpoint → test send), zero-downtime key rotation, and debugging a failed delivery through the DLQ.

2. MCP or REST directly

If you'd rather wire it up yourself:

  • MCP — connect the Model Context Protocol server at mcp.echorelay.dev (JSON-RPC over POST /). initialize, ping and tools/list are public; tools/call needs a management Bearer token.
  • REST — the same operations over plain HTTP at https://api.echorelay.dev, described by a machine-readable OpenAPI at /openapi/management.json.

Both authenticate with a project er_mcp_… Bearer token you create in the panel or mint via the API. (That's different from the er_live_/er_test_ relay keys your senders use against the data plane — see the skill's reference/auth.md for the full model.)

Machine-readable discovery

Everything an agent needs to find our surfaces is published under /.well-known/: an RFC 9727 API catalog, an MCP server card, and the Agent Skills index above. Our homepage also serves a Markdown view to clients that ask for Accept: text/markdown.

Plans, limits and pricing

Agent-driven usage is billed exactly like any other traffic. Rather than restate the numbers here (they change), fetch them live — list_subscription_plans / get_billing via MCP, or GET /billing/plans / GET /billing via REST — or see the pricing page.