What is MCP?
Model Context Protocol is the open spec Anthropic introduced for letting LLM-powered tools discover and use external resources in a standard way. Clients connect to an SSE endpoint, receive a tool manifest, and dispatch tool calls. MCP is what makes Claude Code, Codex, and Gemini CLI composable with third-party systems.
Endpoint
The SSE endpoint is mounted at /mcp/sse on the backend:
Local dev: http://localhost:8000/mcp/sse
Production: https://api.agntdot.com/mcp/sseUsing it with Claude Code
Add AGNT as an MCP server in your Claude Code config:
{
"mcpServers": {
"agnt": {
"url": "https://api.agntdot.com/mcp/sse",
"transport": "sse"
}
}
}After restart, Claude Code sees the AGNT tools in its manifest and can call them directly from inside any chat.
Authentication
MCP clients authenticate the same way external A2A agents do: with an API key Bearer token. Generate one from the developer portal at /developers (authenticated). The token goes in the Authorization header on the SSE handshake.
Tools exposed
The MCP server re-exposes a subset of the A2A protocol as MCP tools. The mapping is 1:1 with the intents documented at A2A protocol:
| MCP tool | A2A intent | Tier |
|---|---|---|
agnt.ping | network.ping | free+ |
agnt.search_venues | booking.search | free+ |
agnt.list_venues | venue.list | free+ |
agnt.confirm_booking | booking.confirm | pro / enterprise |
Related
- A2A protocol— the full envelope schema the MCP tools wrap around
- ClawPulse gateway— the underlying message router
agnt-backend/app/mcp/agnt_network_server.py— the server implementation