ACP — Agent Communication Protocol
ACP is the lightweight HTTP-native cousin of A2A. Where A2A leans on signed agent cards and multi-protocol bindings, ACP focuses on lightweight, language-agnostic agent-to-agent messaging.
Interesting and early. Worth a spike or exploration session.
Agent protocol·Open-source
What It Is
Agent Communication Protocol (ACP) is an open standard for AI agent communication, introduced by IBM's BeeAI project and now under Linux Foundation governance. ACP is lightweight, HTTP-native, and uses JSON-RPC over HTTP or WebSockets, making it easy to integrate into production environments. Agents using ACP can discover, authenticate, and cooperate without bespoke glue code, regardless of the framework or programming language they're built in.
Why It Matters
ACP overlaps significantly with A2A. Both are inter-agent protocols under Linux Foundation governance. The differentiation is design philosophy: ACP is more opinionated about lightweight HTTP and JSON-RPC, while A2A supports multiple bindings (JSON-RPC plus gRPC) and signed agent cards. For AEO buyers, the practical question is whether a customer's AI assistant uses ACP or A2A to reach your business's agent. The reasonable answer is "probably both will exist", with the protocol stack settling into a multi-standard reality similar to how SOAP and REST coexisted for years.
Recent specification work added Trajectory Metadata (tracking multi-step reasoning), Distributed Sessions (URI-based resource sharing across server instances), Citation Metadata (source tracking through agent interactions), and High Availability Support (centralised storage). The BeeAI Framework provides production-grade ACP implementations in Python and TypeScript.
Key Developments
- 2026: Trajectory Metadata, Citation Metadata, Distributed Sessions, and High Availability Support added to the ACP specification.
- 2025: ACP placed under Linux Foundation governance alongside A2A. BeeAI Framework matured for production use.
- 2024: ACP introduced by IBM's BeeAI project.
What to Watch
Watch how ACP and A2A interoperate. Both protocols are under Linux Foundation governance and target similar use cases. Convergence (or stable coexistence) is the realistic outcome. Track BeeAI Framework adoption. The reference implementation matters because protocol adoption is gated by tooling. Watch enterprise buyer preferences. ACP's HTTP-native simplicity may appeal to teams that find A2A's signed-card and multi-binding model heavier than they need.
Strengths
- Lightweight HTTP-native: Easy to integrate into existing production environments.
- Linux Foundation governance: Vendor-neutral, alongside A2A.
- Production framework: BeeAI Framework provides reference implementations in Python and TypeScript.
- Citation Metadata in spec: Direct AEO relevance — agent-to-agent interactions can carry source attribution.
Considerations
- Overlap with A2A: Both target inter-agent communication. Long-term differentiation is unclear.
- Less production traction than A2A: A2A claims 150+ organisations and live deployments at Microsoft, AWS, Salesforce, SAP, ServiceNow.
- Single-vendor origin: IBM-led origin shapes design choices. Cross-vendor adoption is the test of whether it transcends.
- Trust layer immature: No equivalent of A2A's Signed Agent Cards yet, which limits adoption in zero-trust environments.
ACP — Agent Communication Protocol· A2A Protocol· Computer Use· Model Context Protocol· Agent discovery manifests· OAGP — Open Agent Gateway Protocol