api-reference
MCP Protocol
Overview

MCP Protocol Overview

The Model Context Protocol (MCP) is a standardized way for AI assistants to discover and use tools.

What is MCP?

MCP is a JSON-RPC 2.0 based protocol that allows:

  • Tool Discovery: List available tools and their schemas
  • Tool Execution: Call tools with structured arguments
  • Standardized Format: Works with any MCP-compatible client

Protocol Flow

  1. Initialize - Establish connection and negotiate capabilities
  2. List Tools - Discover available tools
  3. Call Tool - Execute tools with arguments
  4. Handle Results - Process structured responses

Endpoints

  • POST /api/v1/atlas/mcp — Public Atlas MCP (four tools: describe_atlas, search_sensors, get_readings, describe_sensor). Prefer this for product integrations and the sensor catalog. See Atlas public MCP and the repo’s docs/MCP_ATLAS_PUBLIC_ENDPOINT.md.
  • POST /api/v1/services/:id/mcp - Service-level MCP endpoint
  • POST /api/v1/sensors/:id/mcp - Sensor-specific endpoint
  • POST /api/v1/projects/:id/mcp - Project-specific endpoint
  • POST /api/v1/sensors/registry/mcp — Legacy registry tool names (same data plane as some Atlas tools; new clients should use /api/v1/atlas/mcp)

JSON-RPC 2.0 Format

All MCP messages follow JSON-RPC 2.0:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/list",
  "params": {}
}

Next Steps

InitializationProtocol