This section introduces the key mechanisms through which Coral operates:

1. Agents are Coralized

Agents are onboarded into the Coral ecosystem through a process called coralization, where they are wrapped with a shared interface that defines their identity, tools, and memory scope. This allows previously standalone or tool-specific code to become an interoperable part of a wider network of agents.

2. Agents Register with the Coral Server

Each Coralized Agent is registered with a Coral Server. This server manages the communication layer including agent discovery, message routing, and thread lifecycle. The server enforces participation rules, scoped access, and structured logging.

3. Agents Collaborate Through Threads

All interactions between agents (and between agents and users) take place in communication threads. A thread is a context-bound space for message exchange, memory sharing, and collaborative task execution.

4. Communication Uses MCP (Model Context Protocol)

All communication in Coral follows the Model Context Protocol (MCP). MCP standardises how messages are structured, how tools are declared and invoked, and how memory is handled.

Each MCP message includes:

Sender and receiver IDs

Thread context

Tool call

Memory and permission metadata

This format ensures traceability, security, and interoperability.

5. Agents Collaborate by Delegating and Invoking Tools

Agents in Coral are not built to handle entire workflows alone. Instead, they mention other agents, invoke tools, and delegate sub-tasks within the structure of the thread.

Coral Protocol brings modular agents into structured collaboration turning isolated reasoning systems into a dynamic society of agents capable of secure, decentralised, and interoperable computation.