Overview
The protocol is fully specified and an SDK exists, plus pilot tests. Results are promising (latency, scalability), but evidence is limited to author-run pilots and lacks independent benchmarks and large-scale production deployments.
Citations0
Evidence Strength0.50
Confidence0.70
Risk Signals11
Trust Signals
Findings with numeric evidence: 4/4
Findings with evidence refs: 4/4
Results with explicit delta: 3/4
Reproducibility
Status: Partial assets available
Open source: Partial
At A Glance
Cost impact: 60%
Production readiness: 60%
Novelty: 70%
Why It Matters For Business
ACP lets organizations stitch together agents from different vendors into secure, auditable workflows. That reduces manual coordination and enables cross-company automation while preserving regulatory controls.
Who Should Care
Summary TLDR
ACP is a multi-layer protocol for direct agent-to-agent (A2A) collaboration across platforms. It combines decentralized discovery, standardized capability descriptions ('Agent Cards'), a negotiation layer for dynamic SLAs, and zero-trust security (DIDs + verifiable credentials). In experiments ACP yields significantly lower latency than raw HTTPS RPC and keeps sub-100 ms latency at 500+ agents, trading some overhead for cross-vendor security and discovery.
Problem Statement
Agents live in isolated stacks and lack a secure, standard way to discover, negotiate with, and reliably execute work with agents on other platforms. That blocks cross-vendor, decentralized workflows.
Main Contribution
ACP specification: a four-layer protocol (Transport, Semantic, Negotiation, Governance/Security) for A2A interaction.
Federated discovery and negotiation: hybrid local/global discovery, Agent Cards for capability advertising, and a four-step negotiation lifecycle (Inquiry→Proposal→Agreement→Execution).
Key Findings
ACP reduces inter-agent communication latency compared to plain JSON-RPC over HTTPS (author claim).
ACP keeps sub-100 ms round-trip latency as agent swarm size grows beyond 500.
Results
| Metric | Value | Baseline | Delta | Split / Dataset | Evidence | Evidence Ref |
|---|---|---|---|---|---|---|
| avg_latency_ms | 58 ms | JSON-RPC over HTTPS 145 ms | ≈60% lower vs JSON-RPC (author abstract states 40% reduction) | Table 3, high-load test | ACP avg latency reported as 58 ms; JSON-RPC 145 ms; MCP 22 ms | Table 3 |
| header_overhead_percent | 8% | MCP 5%, JSON-RPC 12% | slightly higher than MCP, lower than raw JSON-RPC | Table 3 | Table 3 header overhead values | Table 3 |
What To Try In 7 Days
Run the ACP-SDK example and register a test Agent Card to a dev registry.
Create a minimal two-agent workflow (requester + provider) and measure end-to-end latency and success rate.
Compare ACP vs local MCP for a small task to quantify tradeoffs in latency and security.
Agent Features
Memory
Planning
Tool Use
Frameworks
Is Agentic
Yes
Architectures
Collaboration
Optimization Features
Infra Optimization
System Optimization
Reproducibility
Code URLs
Risks & Boundaries
Limitations
Evaluation is limited to author-run pilots; no independent or long-term deployments reported.
Semantic layer may struggle with ambiguous human intents as agents get more complex.
When Not To Use
Within a single trusted application stack where MCP already gives lower latency.
Latency-critical microsecond workloads; ACP targets networked federated agents, not in-process calls.
Failure Modes
Compromised DID/private key leading to impersonation if key management is weak.
Reputation ledger manipulation or Sybil attacks if consortium controls are weak.