NANDA, A2A, and MCP: Complementary Layers of the Agentic Stack
Google's A2A defines how agents talk. Anthropic's MCP connects agents to tools. NANDA's Index helps agents find and trust each other. Here's how they fit together.
Three Protocols, One Stack
The agentic AI ecosystem has coalesced around three foundational protocols, each solving a distinct problem. Understanding their roles — and their boundaries — is essential for anyone building or deploying AI agents at scale.
- Model Context Protocol (MCP) — Created by Anthropic in November 2024 and donated to the Agentic AI Foundation (Linux Foundation) in December 2025. MCP standardizes how AI models connect to external tools, data sources, and services. Think of it as a USB-C port for AI — a universal interface between a model and its environment.
- Agent-to-Agent Protocol (A2A) — Created by Google in April 2025 and donated to the Linux Foundation in June 2025. A2A provides the standardized wire protocol — JSON-RPC 2.0 with Server-Sent Events — for two agents to communicate, negotiate tasks, and stream results. It's the language agents speak to collaborate.
- NANDA Index — Developed by MIT Media Lab's Project NANDA. The NANDA Index provides decentralized agent discovery, identity verification, and trust infrastructure. It's the address book and trust registry that helps agents find each other and decide who to work with.
The Stack: Semantics vs. Syntax
The most clarifying way to understand the relationship between these protocols is through the semantics versus syntax distinction:
Syntax Layer — A2A & MCP
Provides the structural framework: message formats, JSON-RPC protocols, tool interfaces, and technical specifications for data exchange. Answers: "How do agents exchange information?"
Semantics Layer — NANDA Index
Focuses on meaning, context, and intent: agent identity, capability discovery, trust verification, and privacy-preserving resolution. Answers: "How do agents find and trust each other?"
This isn't a hierarchy — both layers are essential. A2A and MCP provide immediate developer tools, code libraries, and security frameworks for building agent communication. NANDA drives the longer-term research into decentralized discovery, privacy-preserving lookups, and federated index architectures.
Interactive · The Agentic Protocol Stack
Not a hierarchy — complementary layers. A2A and MCP provide immediate developer tools and security frameworks. NANDA drives longer-term research into decentralized discovery, privacy-preserving lookups, and federated index architectures.
Click any layer to expand · Each protocol solves a distinct problem in the agentic stack
Feature Comparison
The differences become concrete when you compare capabilities side by side:
| Feature | NANDA Index | A2A | MCP |
|---|---|---|---|
| Focus | Discovery & trust | Agent communication | Tool & data access |
| Discovery | Decentralized Quilt index | Agent Cards at well-known URLs | Server-side tool registration |
| Protocol | Bridges A2A, MCP, HTTPS, gRPC | JSON-RPC 2.0 + SSE | JSON-RPC over stdio/HTTP |
| Identity | Signed AgentAddr + VCs | HTTPS + OAuth/mTLS | Server authentication |
| Privacy | Dual-path resolution | No privacy layer | No privacy layer |
| Governance | MIT Media Lab | Linux Foundation | Agentic AI Foundation (LF) |
The City Analogy
The NANDA team uses a city analogy that makes the relationship intuitive:
- MCP is the plumbing and electrical wiring — it connects AI models to the tools, databases, and services they need to function. Without MCP, agents can't access the resources that make them useful.
- A2A is the road network — it provides the streets and traffic rules that agents use to communicate with each other. Without A2A, agents speak incompatible dialects.
- NANDA is the entire city infrastructure — the address system, the business directory, the trust registry, the licensing authority. Without NANDA, agents can communicate but can't find each other, and have no way to verify who they're talking to.
Each layer is essential. An agent that can talk (A2A) and access tools (MCP) but can't be discovered or trusted is like a business with a phone and a warehouse but no address, no license, and no listing in the directory.
Governance & Convergence
A significant development in the agentic ecosystem has been the convergence of governance. A2A and MCP now operate under the Linux Foundation umbrella, while NANDA remains governed by MIT Media Lab:
- A2A was donated by Google to the Linux Foundation in June 2025, with founding members including Cisco, Salesforce, Intuit, and Atlassian.
- MCP was donated by Anthropic to the Agentic AI Foundation (a directed fund under the Linux Foundation) in December 2025.
- NANDA has engaged with A2A stakeholders including Google, Cisco, and Salesforce from its inception, and the NANDA team has published research specifically designed for compatibility with both A2A and MCP.
This governance convergence is not accidental. The Linux Foundation recognized early that fragmented agent infrastructure would slow adoption, and has actively worked to bring complementary protocols under a shared governance umbrella. Cisco's AGNTCY project (July 2025) further bridges these efforts by making A2A agents and MCP servers discoverable through shared agent directories.
What NANDA Adds That A2A and MCP Don't
A2A explicitly describes several concerns as "ancillary" that the NANDA Index treats as essential infrastructure:
- Global agent discovery — A2A's Agent Cards sit at
/.well-known/agent.jsonon individual domains. There's no global index. The NANDA Quilt provides decentralized discovery across the entire ecosystem. - Identity and credentials — A2A relies on standard HTTPS and optional OAuth/mTLS. NANDA provides cryptographically signed AgentFacts with W3C Verifiable Credentials, enabling fine-grained capability attestations.
- Economic primitives — NANDA's Phase 2 roadmap includes knowledge pricing, resource markets, and payment protocols for agentic commerce — capabilities neither A2A nor MCP address.
- Reputation and trust — Beyond authentication, NANDA tracks behavioral history and trust scores, enabling agents to make informed delegation decisions.
- Privacy-preserving lookups — NANDA's dual-path resolution enables anonymous agent discovery through neutral third-party hosts — critical for sensitive industries like healthcare and finance.
From the A2A Team
"We're proud to support Project Nanda's work as they utilize the Agent2Agent protocol for their advanced research on the internet of agents. Our goal with A2A is to empower businesses, consumers, and researchers with an open standard that enables the innovation needed to solve complex challenges that are beyond the scope of any single agent."
— Rao Surapaneni, VP, Google Cloud
This endorsement underscores the complementary nature of the relationship: Google's A2A team explicitly recognizes that NANDA's discovery and trust infrastructure extends the reach of A2A into problems that A2A wasn't designed to solve.
Building With the Full Stack
For developers and organizations deploying AI agents today, the practical implications are clear:
- Use MCP to connect your agents to tools, databases, and external services
- Use A2A to enable agent-to-agent communication and task delegation
- Use the NANDA Index to make your agents globally discoverable, verifiable, and trustworthy
The NANDA Adapter bridges all three, enabling cross-protocol interoperability so agents built on any protocol can discover and work with agents on any other. An MCP assistant can discover an A2A inventory agent through the NANDA Index and communicate seamlessly through the NANDA Adapter.