Engineering brief
ACP: The protocol that could finally decouple clients from agent harnesses
At a glance
- Relevance
- Practical value
- Warnings
- None
ACP aims to do for agent orchestration what MCP did for tool connections. If it gains adoption, your team could swap client UIs without changing agent harnesses.
Standards create ecosystems. ACP could do for agent orchestration what MCP did for agent tooling.
Summary
Most AI agent harnesses today use custom or bespoke client interfaces, creating a fragmented ecosystem where one client can only control one harness. Alex Hancock from Block proposes ACP (Agent Client Protocol), a universal standard that decouples client applications from agent harnesses. The analogy is clear: MCP standardized how agents connect to tools; ACP
standardizes how clients connect to agents. ACP uses JSON-RPC messages for sessions, user messages, agent responses, tool call notifications, and permission requests. It supports both local (stdio) and remote (HTTP/WebSocket) transports. Hancock demonstrated two different clients (Zed editor and a terminal client) controlling the same Goose agent over ACP, then showed a custom Python
client connecting remotely to the same agent harness. The real organizational impact is architectural flexibility. Teams can now independently move client, harness, tool, and model components across machines or clouds. This means engineering leaders can choose the best client UX for their team without being locked into a specific harness or tool ecosystem.
ACP's extensibility via custom methods allows emergent standardization. Tradeoffs exist: ACP is still early, with few production adoptions beyond the initial editor companies (Zed, JetBrains) and Goose. Remote transport just landed. The standard requires community buy-in to reach its potential. Engineers should watch whether ACP gains MCP-like adoption or remains a niche protocol.
Watch the video
This video is blocked due to your privacy settings. To watch this video, please accept YouTube marketing cookies.
Related breakdowns
AI agents fail without organizational context: the case for context engineering
AI agents are smart but ignorant of your organization's history. Context engineering solves the gap between code that compiles and code that works.
LLM inference is a memory problem, not a compute problem
Inference cost is the hidden operational tax on AI products. This workshop breaks down the KV cache bottleneck, model vs. serving optimisations, and when VLM…
X402 Isn't Ready: The State of Agentic Payments in 2025
Agentic payments are hyped, but X402 still has unresolved double-spending, conflicting status codes with MCP, and requires clunky workarounds. Read why Apify…
Get TL;DW
Too Long; Didn't Watch.
A concise breakdowns of the AI and devtools videos that actually matter for engineering leaders.
Free. Weekly. No hype.
Video and thumbnails remain the property of their respective creators. tldw.news provides editorial analysis, commentary, and discovery links to original content.