We want to introduce an ACP (Agent Client Protocol) layer to connect Claude Code with our existing agent system.
The goal is not to rewrite the architecture, but to add a clean interface between the editor and our orchestrator.
Right now, our system already works:
- agents collaborate
- tools/MCPs execute tasks
- workflows run internally
This change should not break or replace any of that.
What we are adding
A new component:
dev-orchestrator` (ACP agent)
This will:
- receive tasks from Claude Code
- forward them to our existing orchestrator
- return progress and results back to the editor
Important (what this is NOT)
- This is not replacing MCP
- This is not replacing our orchestrator
- This is not moving logic into ACP
ACP is just:
a communication layer between Claude Code and our system
Where it fits
Claude Code
↓
ACP (dev-orchestrator)
↓
Existing orchestrator (Agno / LangGraph / etc)
↓
MCP tools (dotcontext + others)
Why this improves things
- cleaner separation between editor and backend logic
- easier to plug into Claude Code
- keeps our current system intact
- allows better session handling and visibility
- makes future scaling easier (more agents, more tools)
Scope (MVP)
-
create dev-orchestrator ACP agent
-
support basic ACP flow:
- initialize
- session/new
- session/prompt
- session/update
- session/cancel
-
connect to existing orchestrator
-
add adapter to dotcontext MCP (for context + workflow)
-
read workflow state before execution
-
update workflow after execution
Acceptance Criteria
- Claude Code can start a session with
dev-orchestrator
- tasks flow through ACP → orchestrator → MCP tools
- current agent workflows still work as before
- dotcontext workflow (PREVC) is respected
- no regression in existing behavior
Notes
This is meant to be a thin layer, not a rewrite.
If done correctly:
- nothing breaks
- everything becomes more structured
- future integrations become easier
https://agentclientprotocol.com/get-started/architecture
npm install @agentclientprotocol/sdk
import { AgentSideConnection } from "@agentclientprotocol/sdk";
Agent stdio
const agentConnection = new AgentSideConnection(
(conn) => new MyCustomAgent(conn),
process.stdin,
process.stdout,
);

We want to introduce an ACP (Agent Client Protocol) layer to connect Claude Code with our existing agent system.
The goal is not to rewrite the architecture, but to add a clean interface between the editor and our orchestrator.
Right now, our system already works:
This change should not break or replace any of that.
What we are adding
A new component:
dev-orchestrator` (ACP agent)
This will:
Important (what this is NOT)
ACP is just:
Where it fits
Claude Code
↓
ACP (dev-orchestrator)
↓
Existing orchestrator (Agno / LangGraph / etc)
↓
MCP tools (dotcontext + others)
Why this improves things
Scope (MVP)
create
dev-orchestratorACP agentsupport basic ACP flow:
connect to existing orchestrator
add adapter to dotcontext MCP (for context + workflow)
read workflow state before execution
update workflow after execution
Acceptance Criteria
dev-orchestratorNotes
This is meant to be a thin layer, not a rewrite.
If done correctly:
https://agentclientprotocol.com/get-started/architecture
npm install @agentclientprotocol/sdk
import { AgentSideConnection } from "@agentclientprotocol/sdk";
Agent stdio
const agentConnection = new AgentSideConnection(
(conn) => new MyCustomAgent(conn),
process.stdin,
process.stdout,
);