# Agent2Agent Protocol (A2A) vs Natural Language Interaction Protocol (NLIP)

The map lists them as complementary: sites and agents commonly implement both.

[6 Coordinate](/category/coordinate/) [Agent2Agent Protocol](/protocols/a2a/) · [6 Coordinate](/category/coordinate/) [Natural Language Interaction Protocol](/protocols/nlip/)

|  | [Agent2Agent Protocol](/protocols/a2a/) A2A | [Natural Language Interaction Protocol](/protocols/nlip/) NLIP |
| --- | --- | --- |
| Lifecycle step | 6 Coordinate: Talk to another agent | 6 Coordinate: Talk to another agent |
| Answers | How do two independent agents, built on different frameworks, discover each other and delegate a task to completion? | Can this agent or service exchange natural-language, multi-modal messages using a shared, vendor-neutral envelope instead of a bespoke API? |
| Governed by | Linux Foundation (Technical Steering Committee: AWS, Cisco, Google, IBM Research, Microsoft, Salesforce, SAP, ServiceNow); Agentic AI Foundation (AAIF) since 17 August 2026 | Ecma International Technical Committee TC56 (NLIP), with an ISO ballot in progress as of mid-2026 per contributor reports; incubated and supported by the Enterprise Neurosystem community |
| Status (issuer's words) | Stable, v1.0.1 | Ecma International Standard, 1st edition, published 10 December 2025 (ECMA-430 core spec, plus companion ECMA-431/432/433 transport bindings, ECMA-434 security profiles, and ECMA TR/113 explanatory report); an ISO standardisation ballot was reported in progress by project contributors as of August 2026 |
| Version | 1.0.1 (2026-05-28) | ECMA-430, 1st edition (2025-12-10) |
| Form / transport | endpoint | document format |
| Found at | /.well-known/agent-card.json | none |
| Implementation | multi-implementation — 150+ supporting organizations, claimed by the Linux Foundation on 9 April 2026 (https://www.linuxfoundation.org/press/a2a-protocol-surpasses-150-organizations-lands-in-major-cloud-platforms-and-sees-enterprise-production-use-in-first-year). Card-version censuses are separate measurements, kept with their frames and not averaged. Agenstry 2026-09-01 (https://agenstry.com/blog/state-of-the-agent-card): of 1,910 cards declaring protocolVersion, 65.5% (1,252) are 0.3.x and 25.5% (488) are 1.0.x. Emissar 2026-10-02 (https://emissar.ai/learn/a2a-adoption-2026): 276 cards on the Tranco top 1M (205 in v1.0 shape, 71 in v0.3 shape, 0 signed, 0 in the top 1,000 domains). APIs.io / Kin Lane 2026-09-21 (https://apievangelist.com/2026/09/21/a2a-adoption-is-421-of-27840-and-most-of-them-arrived-with-the-card/): 421 parseable cards out of 4,718 hosts probed (27,840 providers). rStation 2026-10-12 (https://rstation.ai/problems/ai_agent/papers/1): 661 cards on 2,164 x402 seller hosts (189 pass their attributed version and fail a check that assumes 1.0). | multi-implementation — NLIP project / Enterprise Neurosystem (self-reported: PyPI nlip_sdk/nlip-server downloads ~300-420/month as of 2026-06; production use claimed by Enterprise Neurosystem member Kove; incubation partners CUNY and IBM) |
| Evidence | 11 verified, 2 reported | 5 verified |
| Primary use case | Let agents talk to other agents | Let agents talk to other agents |
| Last verified | 2026-10-23 (current) | 2026-10-23 (current) |
| Specification | [https://a2a-protocol.org/latest/specification/](https://a2a-protocol.org/latest/specification/) | [https://ecma-international.org/publications-and-standards/standards/ecma-430/](https://ecma-international.org/publications-and-standards/standards/ecma-430/) |

## When to use which

- Choose Agent2Agent Protocol when your question is “How do two independent agents, built on different frameworks, discover each other and delegate a task to completion?” and you accept its maturity: status "Stable, v1.0.1", implementation multi-implementation (150+ supporting organizations, claimed by the Linux Foundation on 9 April 2026 (https://www.linuxfoundation.org/press/a2a-protocol-surpasses-150-organizations-lands-in-major-cloud-platforms-and-sees-enterprise-production-use-in-first-year). Card-version censuses are separate measurements, kept with their frames and not averaged. Agenstry 2026-09-01 (https://agenstry.com/blog/state-of-the-agent-card): of 1,910 cards declaring protocolVersion, 65.5% (1,252) are 0.3.x and 25.5% (488) are 1.0.x. Emissar 2026-10-02 (https://emissar.ai/learn/a2a-adoption-2026): 276 cards on the Tranco top 1M (205 in v1.0 shape, 71 in v0.3 shape, 0 signed, 0 in the top 1,000 domains). APIs.io / Kin Lane 2026-09-21 (https://apievangelist.com/2026/09/21/a2a-adoption-is-421-of-27840-and-most-of-them-arrived-with-the-card/): 421 parseable cards out of 4,718 hosts probed (27,840 providers). rStation 2026-10-12 (https://rstation.ai/problems/ai_agent/papers/1): 661 cards on 2,164 x402 seller hosts (189 pass their attributed version and fail a check that assumes 1.0).).
- Choose Natural Language Interaction Protocol when your question is “Can this agent or service exchange natural-language, multi-modal messages using a shared, vendor-neutral envelope instead of a bespoke API?” and you accept its maturity: status "Ecma International Standard, 1st edition, published 10 December 2025 (ECMA-430 core spec, plus companion ECMA-431/432/433 transport bindings, ECMA-434 security profiles, and ECMA TR/113 explanatory report); an ISO standardisation ballot was reported in progress by project contributors as of August 2026", implementation multi-implementation (NLIP project / Enterprise Neurosystem (self-reported: PyPI nlip_sdk/nlip-server downloads ~300-420/month as of 2026-06; production use claimed by Enterprise Neurosystem member Kove; incubation partners CUNY and IBM)).

Guidance is derived from the two records (question, status, implementation, prerequisites, pairings), not from opinion. Each record's page carries its numbered references.

## Read the full articles

- [Agent2Agent Protocol](/protocols/a2a/): How do two independent agents, built on different frameworks, discover each other and delegate a task to completion?
- [Natural Language Interaction Protocol](/protocols/nlip/): Can this agent or service exchange natural-language, multi-modal messages using a shared, vendor-neutral envelope instead of a bespoke API?

Ask the MCP server to compare any two: the compare-protocols prompt, or get on both ids. [Connect](/connect/).

[a2a.json](/data/protocols/a2a.json) · [nlip.json](/data/protocols/nlip.json) · [Markdown](/compare/a2a-vs-nlip/index.md)