MCP integration
Connect AI agent runtimes (Anthropic, OpenAI, custom) to Agent Number via Model Context Protocol. One endpoint, thirteen tools, zero adapters.
Endpoint
MCP server (streamable-http transport):
https://agentnumber.really.com/mcp
Discoverable from our MCP server card at /.well-known/mcp/server-card.json.
Install in Claude Code
One line registers Agent Number as an MCP server in your Claude Code session:
claude mcp add agent-number https://agentnumber.really.com/mcp
Other MCP-aware runtimes — Cursor, Zed, ChatGPT, OpenClaw — point their MCP client at the same URL (streamable-http transport).
Initialize
Standard MCP handshake. The server returns its protocol version + capabilities:
curl -X POST https://agentnumber.really.com/mcp \
-H "content-type: application/json" \
-H "accept: application/json, text/event-stream" \
-d '{
"jsonrpc": "2.0",
"id": "1",
"method": "initialize",
"params": {
"protocolVersion": "2025-11-25",
"clientInfo": {"name": "my-agent", "version": "1.0"}
}
}'
Tools
Call tools/list to discover, then invoke directly. The
full tool catalog with arguments and response shapes lives on the
dedicated Tools page.
Authentication via MCP
Bearer-token tools require Authorization: Bearer <token> on the MCP request itself. Mint a token via subscribe (after payment) or auth (wallet-signature recovery). See Auth.
Try the demo first
Before paying, call try_demo. It's a free, no-auth simulated walkthrough — the response shapes match the real endpoints exactly, so you can scaffold your integration and verify wiring before the real subscribe:
curl -X POST https://agentnumber.really.com/mcp \
-H "content-type: application/json" \
-H "accept: application/json, text/event-stream" \
-d '{
"jsonrpc": "2.0",
"id": "demo",
"method": "tools/call",
"params": {"name": "try_demo", "arguments": {}}
}'
Long-poll behavior (receive_sms)
receive_sms holds the request open up to wait_seconds (default 30, max 120) and emits MCP notifications/progress every 20 seconds to keep the client alive indefinitely. Near-instant OTP delivery end-to-end.
Discovery files for crawlers and agents
/.well-known/mcp/server-card.json— full MCP server card/.well-known/mcp.json— compact MCP descriptor/.well-known/agent-card.json— A2A Agent Card/.well-known/agent-skills/index.json— Anthropic Agent Skills index/agent-skill/SKILL.md— full SKILL.md (digest-verified)