ACP vs ACP: Agent Client Protocol or Agentic Control Plane?
Two unrelated projects share the acronym. dsh acp, which current dsh runs as dsh --profile acp, is the Agent Client Protocol, the JSON-RPC standard Zed introduced for connecting editors to coding agents. The Agentic Control Plane (this site) checks each dsh tool call against your policy and logs the decision. DeepSeek Harness ships both, so search results mix them. The table routes you to the right one.
One caveat on the second project. The dsh plugin currently records and reports decisions. Enforcing deny is being fixed. dsh has gone on to run a denied call in one production workspace, including on a safety-floor deny.
| You’re trying to | You want | Go to |
|---|---|---|
| Use dsh from Zed or another editor | Agent Client Protocol | The community deepseek-harness-acp (full editor vocabulary). dsh’s own dsh --profile acp is the official server, but the dsh CLI README describes it as serving automation clients |
| Drive dsh programmatically from your app | Agent Client Protocol | dsh --profile acp (official) or @ai-sdk/harness-acp (Vercel AI SDK harness layer) |
| Read the protocol spec | Agent Client Protocol | agentclientprotocol.com |
| Run Vercel’s fx under an editor or client | Agent Client Protocol | fx acp, fx’s built-in protocol server (fx control model) |
| See and control what dsh does: allow / ask / deny each tool call, with an audit log | Agentic Control Plane | @agenticcontrolplane/dsh, launch post, dsh control model |
| Agent-to-agent messaging between frameworks | Neither. See the third ACP below | A2A is where that layer settled |
The Agent Client Protocol (the one in dsh core)
Agent Client Protocol is a JSON-RPC standard for connecting code editors and automation clients to coding agents. It does for agents what LSP did for language tools. Zed introduced it in 2025, JetBrains followed, and a registry now lists compatible agents. You get one protocol instead of a custom integration for every editor and agent pair.
dsh supports it in core. The packages/acp directory in the dsh repo is this protocol. Current dsh runs its ACP server as a profile: dsh --profile acp drives harness agents over stdio. The dsh CLI README puts it as “SDK and ACP are profiles, not separate public bins.” The server package is still on npm as @deepseek-ai/dsh-acp, in pre-release versions only, and older write-ups point there.
The community deepseek-harness-acp goes further for editor use. It maps dsh’s session-event log onto the full protocol vocabulary: streamed text and reasoning, tool calls with diffs, plans and permission requests. If that is what you came for, those links are your answer.
The Agentic Control Plane (this site)
The Agentic Control Plane doesn’t connect anything to anything. It sits on dsh’s tools/pre-execute pipeline. For each tool call it decides whether the call should run: allow, ask a human, or deny, against policies you set. Every decision is logged with its reason.
@agenticcontrolplane/dsh is a native plugin of about 270 lines on dsh’s typed interception points. The launch post covers what it intercepts and what it caught on day one.
See /controls/dsh for what the dsh plugin enforces today.
Different layers, used together
The protocol is transport. The control plane is judgment. In principle one dsh session can use both: your editor talks to the agent over the Agent Client Protocol, and the plugin checks tool calls in whichever profile it is installed in. We have not tested the plugin inside dsh’s acp profile. Until the coverage matrix lists it, don’t count on a dsh session driven from an editor being checked or recorded by ACP.
They meet at one point. The protocol defines session/request_permission, where the agent asks the client to approve a tool call with options like allow_once, allow_always, reject_once and reject_always. The spec notes clients may auto-answer based on user settings. Our dsh plugin does not work at that point. It checks calls on dsh’s tools/pre-execute hook, and we have not tested it under an editor.
The third ACP
There is also an Agent Communication Protocol (IBM Research / BeeAI, 2025) for agent-to-agent messaging. In that layer the ecosystem consolidated on A2A, which Google donated to the Linux Foundation. If you searched “ACP” meaning agent-to-agent interop, A2A is likely what you want.
Name-to-package map
| Name | Which ACP | What it is |
|---|---|---|
packages/acp (dsh repo) |
Agent Client Protocol | Protocol implementation in dsh core |
dsh --profile acp |
Agent Client Protocol | Official ACP server for dsh, automation clients, JSON-RPC over stdio |
@deepseek-ai/dsh-acp |
Agent Client Protocol | The same server’s npm package |
deepseek-harness-acp (openma-ai) |
Agent Client Protocol | Community ACP server with the full editor vocabulary |
@ai-sdk/harness-acp |
Agent Client Protocol | Vercel AI SDK adapter for any ACP-compatible harness |
fx acp (Vercel Labs fx) |
Agent Client Protocol | fx’s built-in protocol server. It routes each approval-gated call to the client as session/request_permission, fail-closed (details) |
@agenticcontrolplane/dsh (launch name dsh-plugin-acp) |
Agentic Control Plane | Policy check on each dsh tool call, decisions logged. Deny enforcement is being fixed |
Common questions
deepseek acp: which package do I install?
For an editor, the community deepseek-harness-acp, which carries the full editor vocabulary. For automation clients, dsh’s own acp profile (dsh --profile acp), which npm describes as automation-only. For allow, ask and deny on each dsh tool call with a log, @agenticcontrolplane/dsh, the Agentic Control Plane plugin. The dsh plugin records and reports decisions today. Enforcing deny is being fixed.
agent client protocol compatible tools: which ones does this page cover?
Editors: Zed, which introduced the protocol, and JetBrains. Agents and adapters: DeepSeek Harness through dsh --profile acp and deepseek-harness-acp, Vercel’s fx through fx acp, and the Vercel AI SDK’s @ai-sdk/harness-acp for driving any compatible harness from code. The protocol’s own registry at agentclientprotocol.com is the full list.
Is dsh's ACP the same thing as the Agentic Control Plane?
No. DeepSeek Harness’s packages/acp and its acp profile (dsh –profile acp) implement the Agent Client Protocol, which Zed introduced, a JSON-RPC standard for connecting editors and automation clients to coding agents. The Agentic Control Plane is an unrelated project that shares the acronym. It is a policy layer that checks each tool call dsh makes before it runs (allow, ask or deny) and records every decision. dsh ships both. We have not yet tested the plugin inside dsh’s acp profile.
How do I use DeepSeek Harness with Zed or another editor?
Use the Agent Client Protocol side. The community package deepseek-harness-acp maps dsh’s session events onto the full ACP vocabulary for editors like Zed. dsh’s own acp profile (dsh –profile acp) is the official server, described by DeepSeek as automation-only. The Vercel AI SDK’s @ai-sdk/harness-acp adapter drives any ACP-compatible harness programmatically.
What's the difference between the Agent Client Protocol and the Agent Communication Protocol?
A third collision. The Agent Client Protocol (Zed, 2025) connects editors and clients to coding agents and is widely adopted at that layer. The Agent Communication Protocol (IBM Research / BeeAI, 2025) targets agent-to-agent messaging, a layer where A2A, now a Linux Foundation project, has become the de facto standard.
Can I use the Agent Client Protocol and the Agentic Control Plane together?
They sit at different layers, so in principle yes. The protocol carries the conversation between your editor and the agent, including session/request_permission when the agent wants approval for a tool call. The control plane checks each tool call against your rules and logs the decision. We have not yet tested the dsh plugin in a session driven from an editor, so don’t count on those calls being checked or recorded until the coverage matrix lists it.
Running dsh and want the control side? One command: curl -sf https://agenticcontrolplane.com/install.sh | bash, free up to 5 agents. The living reference for what can be seen and controlled in dsh is at /controls/dsh.