Protocol Overview

Qualixar OS supports two standard protocols that serve different but complementary purposes. Understanding when to use each is key to getting the most out of QOS.

Two Protocols, Two Purposes

MCP (Model Context Protocol)

MCP is a JSON-RPC 2.0 based standard created by Anthropic for connecting LLM applications to external tools and data sources. In QOS, MCP is the tool access layer -- it lets IDEs and AI coding assistants call QOS functions like run_task, search_memory, or list_agents.

When to use MCP:

• Connecting QOS to an IDE (Claude Code, Cursor, VS Code, Windsurf, Cline)
• Exposing QOS orchestration as callable tools for an LLM
• Building custom clients that need to invoke QOS operations
• Any scenario where a human or LLM needs to call QOS functions

Implementation: src/channels/mcp-server.ts (stdio transport, 25 individual tools) and src/commands/adapters/mcp-adapter.ts (6 domain-grouped tools for token efficiency).

A2A (Agent-to-Agent Protocol)

A2A is Google's open protocol for agent-to-agent communication. In QOS, A2A is the inter-agent communication layer -- it lets agents discover each other, delegate tasks, and exchange results without human intervention.

When to use A2A:

• Connecting QOS to other agent frameworks (LangGraph, CrewAI, AutoGen)
• Building multi-system agent workflows where agents in different processes need to collaborate
• Exposing QOS agents as discoverable services on your network
• Any scenario where agents need to talk to agents

Implementation: src/compatibility/a2a-server.ts (inbound), src/compatibility/a2a-client.ts (outbound), src/agents/transport/a2a-transport.ts (inter-agent transport with circuit breaker), src/agents/transport/a2a-msghub.ts (transparent A2A wrapper for the internal message hub).

How They Complement Each Other

                  MCP                          A2A
            (Tool Access)              (Agent Communication)
                 |                              |
    IDE / LLM Client                   External Agent System
         |                                      |
    [JSON-RPC 2.0]                    [HTTP + Agent Cards]
         |                                      |
    QOS MCP Server                     QOS A2A Server
         |                                      |
         +----------> Orchestrator <------------+
                          |
                    Internal Agents
                    (MsgHub + A2A)

An IDE calls QOS via MCP to submit a task. The orchestrator assigns that task to internal agents. Those agents communicate internally using A2A message format via the MsgHub. If the task requires capabilities from an external agent, QOS discovers and delegates to it via A2A.

The two protocols never conflict. MCP handles the vertical connection (IDE down to QOS). A2A handles the horizontal connection (QOS across to other agents).

Protocol Versions

Protocol	Version	Specification
MCP	Latest	modelcontextprotocol.io
A2A	v0.3	google.github.io/A2A

QOS validates the A2A protocol version strictly -- agent cards must declare protocol: 'a2a/v0.3' or discovery will fail (enforced in src/compatibility/a2a-client.ts, line 254).

Quick Comparison

Dimension	MCP	A2A
Purpose	Tool access for LLMs/IDEs	Agent-to-agent communication
Wire format	JSON-RPC 2.0	HTTP REST + JSON
Discovery	Client configuration	/.well-known/agent-card endpoint
Direction	Client calls server	Bidirectional (client + server)
Transport	stdio, SSE, WebSocket	HTTP
Authentication	QOS_API_KEY env var	Per-agent (future)
Statefulness	Stateless tool calls	Stateful task lifecycle

Next Steps

• MCP Protocol Details -- Full MCP implementation, all 25 tools, connecting from any client
• A2A Protocol Details -- A2A server/client, agent card discovery, inter-agent transport