AmpersandBoxDesign AmpersandBoxDesign/docs/comparison-table.md
This document positions the [&] Protocol relative to adjacent standards, research efforts, and ecosystem protocols.

[&] vs MCP vs A2A vs DALIA vs ACP

This document positions the [&] Protocol relative to adjacent standards, research efforts, and ecosystem protocols.

The short version:

  • MCP standardizes agent-to-tool connectivity.

  • A2A standardizes agent-to-agent coordination.

  • ACP standardizes lightweight agent communication over network APIs.

  • DALIA highlights the need for compositional intelligence architectures at the research level.

  • [&] defines the capability composition layer: how an agent declares, validates, binds, and governs its cognitive capabilities before compiling into runtime protocols.

The right framing is not replacement, but composition:

[&] compiles into MCP + A2A configurations.

Comparison Table

SystemPrimary ScopeLayer in the StackMain Unit of AbstractionWhat It Standardizes WellWhat It Does Not StandardizeRelationship to [&]
[&] ProtocolCapability composition for agentsCompositionCapability declarations, contracts, provenance, governanceHow memory, reasoning, time, space, and governance capabilities are declared, validated, composed, and compiled into downstream configsLow-level transport protocols, direct tool RPC semantics, agent network communicationSource-of-truth layer
MCPAgent-to-tool connectivityContext / ToolingTools, resources, prompts, client-server interactionTool discovery, tool invocation, structured context access, resource exposureHow an agent's cognitive capabilities are selected, composed, governed, or provenance-linked[&] can generate MCP configuration
A2AAgent-to-agent coordinationCoordinationAgent cards, skills, tasks, delegation flowsAgent discovery, advertised skills, delegation and inter-agent interactionHow internal cognitive capabilities are composed or validated before publication[&] can generate A2A agent cards
ACPLightweight agent communicationCommunicationHTTP/REST-style agent endpoints and message exchangeInter-agent messaging and integration patterns over standard web infrastructureCapability algebra, provenance semantics, governance-as-data, typed capability compositionAdjacent network protocol; [&] sits at a different concern boundary
DALIAResearch framework for compositional intelligenceResearch / TheoryCognitive architecture modelWhy modular, compositional agent intelligence matters; academic framing for layered capability systemsA production-ready protocol artifact, schema, CLI, registry, or runtime config formatIndependent conceptual validation of the gap [&] addresses

Quick Definitions

[&] Protocol

An open, language-agnostic protocol for composing agent cognitive capabilities. It defines:

  • canonical ampersand.json

  • capability primitives like &memory, &reason, &time, &space, and &govern

  • namespaced subtypes like &memory.graph and &time.anomaly

  • capability contracts with typed operations

  • composition algebra properties

  • provenance records

  • governance constraints

  • compilation targets such as MCP and A2A artifacts

MCP

The Model Context Protocol focuses on connecting agents to tools, resources, and prompts. It standardizes the agent-to-tool interface, not the internal capability model of the agent.

A2A

The Agent-to-Agent Protocol focuses on how agents discover each other, advertise skills, and coordinate tasks. It assumes an agent already exists and exposes an interaction surface.

ACP

The Agent Communication Protocol focuses on interoperable communication between agents or agent-like services, often through lightweight network-native interfaces.

DALIA

A research effort that treats intelligence as compositional and layered. DALIA helps articulate why capability composition matters, but it is not itself a deployable protocol specification in the same sense.

The Stack View

A useful mental model is:

LayerQuestionExample Protocols
UIHow does the agent render and interact with users?AG-UI, A2UI
CompositionWhat capabilities make up this agent, and how do they fit together?[&]
CoordinationHow do agents discover and delegate to each other?A2A
ContextHow does an agent access tools and resources?MCP
CommunicationHow do services and agents exchange messages over networks?ACP, HTTP, gRPC

This is why [&] should not be framed as "above" MCP or A2A in a political or replacement sense. A better and more accurate statement is:

[&] defines the composition layer and compiles into MCP + A2A configurations.

Where MCP Is Strong

MCP is strong when you need:

  • a standard way to expose tools

  • structured resources and prompts

  • agent access to external systems

  • clear client/server boundaries

  • tool interoperability across runtimes

MCP is not trying to answer questions like:

  • Should this agent have episodic memory or graph memory?

  • Is &time.anomaly compatible with &reason.argument in this pipeline?

  • What governance rules constrain this agent's actions?

  • How do I preserve a provenance chain across composed cognition?

Those are composition-layer concerns, which is exactly why [&] exists.

Where A2A Is Strong

A2A is strong when you need:

  • agent discovery

  • advertised skills

  • task delegation

  • multi-agent workflow handoff

  • machine-readable agent cards

A2A starts from the perspective that an agent has already been built and is now publishing its capabilities outward.

[&] starts earlier. It defines how the agent is composed in the first place.

That makes the relationship complementary:

  • [&] describes internal composition

  • A2A describes external coordination

Where ACP Is Strong

ACP is strong when you need:

  • lightweight inter-agent communication

  • simple network-native interfaces

  • integration over common web patterns

  • deployable communication surfaces without requiring tool-specific semantics

ACP is closest to a communications substrate. It is useful for transport and interoperability, but it does not define:

  • capability primitives

  • capability subtypes

  • composition algebra

  • provenance chain structure

  • governance constraints as portable data

So ACP and [&] solve different problems.

Why DALIA Matters Here

DALIA is important because it strengthens the claim that there is a real conceptual gap in the ecosystem.

It supports ideas like:

  • intelligence as composition rather than monolith

  • modular cognitive systems

  • layered capabilities

  • architectural clarity between memory, reasoning, planning, governance, and execution

That is useful because [&] is not just inventing a product category. It is operationalizing a real architectural need into:

  • schema artifacts

  • machine-readable contracts

  • validation logic

  • generation tools

  • protocol-ready outputs

DALIA is a research signal; [&] is a protocol implementation strategy.

Decision Guide

If your question is:

"How does my agent call tools?"

Use MCP.

"How do my agents find and delegate to each other?"

Use A2A.

"How do my agent services communicate over network APIs?"

Use ACP.

"How should I think about intelligence as a compositional architecture?"

Read DALIA.

"How do I declare, validate, govern, and compile an agent's cognitive capabilities?"

Use [&].

Side-by-Side by Concern

Concern[&]MCPA2AACPDALIA
Capability declarationYesNoNoNoConceptually
Capability composition algebraYesNoNoNoConceptually
Typed capability contractsYesPartial at tool boundarySkill-level onlyNoConceptually
Governance as portable dataYesNoLimited / indirectNoConceptually
Provenance chain semanticsYesNot coreNot coreNoConceptually
Tool invocation protocolNoYesNoNoNo
Agent discovery / delegationNoNoYesPartialNo
Network communication substrateNoPartialPartialYesNo
Schema-first source of truthYesYes, but for tool interfacesYes, for agent cardsVariesNo
Runtime configuration generationYesN/AN/AN/ANo

The Important Non-Competitive Framing

This repository should consistently avoid the claim that [&] competes with MCP or A2A.

The accurate position is:

  • MCP and A2A are valuable standards

  • they solve different layers of the stack

  • neither standardizes internal capability composition

  • [&] fills that missing concern

  • [&] produces artifacts that are useful to MCP and A2A ecosystems

The safest one-line framing is:

[&] is the source-of-truth composition layer that compiles into MCP + A2A configurations.

Practical Example

A developer may start with this declaration:

  • &memory.graph

  • &time.anomaly

  • &space.fleet

  • &reason.argument

  • &govern.escalation

From that, [&] can produce:

  1. a validated ampersand.json

  2. composition checks and compatibility validation

  3. governance metadata

  4. provenance requirements

  5. an MCP config for tool-facing runtime integration

  6. an A2A agent card for coordination-facing publication

MCP alone does not give you the declaration model. A2A alone does not give you the declaration model. ACP alone does not give you the declaration model.

That declaration model is the missing layer.

Summary

The five systems are best understood as different answers to different questions:

  • MCP: how an agent uses tools

  • A2A: how agents work with other agents

  • ACP: how agents communicate over APIs

  • DALIA: why compositional intelligence matters

  • [&]: how an agent's capabilities are declared, composed, validated, governed, and compiled

That distinction is the core positioning of the [&] Protocol.

Open in the interactive atlas