edgecrab-acp

Why this crate? VS Code (and other IDEs) speak Agent Communication Protocol
— a JSON-RPC 2.0 stdio protocol that lets editors embed agents without shipping their own
LLM infrastructure. edgecrab-acp wraps the full EdgeCrab agent as an ACP server, so
you get EdgeCrab's 30+ tools, persistent memory, and multi-provider LLM support directly
inside your editor sidebar — no browser, no separate terminal.

Part of EdgeCrab — the Rust SuperAgent.

Start the ACP server

edgecrab acp

The process communicates over stdin/stdout using JSON-RPC 2.0. Your IDE connects to
the process directly — nothing to configure beyond pointing the IDE at the binary.

VS Code quick setup

Install the EdgeCrab VS Code extension (or add a manual ACP entry):

// .vscode/settings.json
{
  "acp.agents": [
    {
      "id": "edgecrab",
      "name": "EdgeCrab",
      "command": "edgecrab",
      "args": ["acp"]
    }
  ]
}

Open the Agent panel → select EdgeCrab → start chatting.

Protocol flow

IDE            edgecrab-acp              edgecrab-core
 │─ agent/run ──────────────────────────────▶│
 │              │─ Agent::run_conversation() ─▶│
 │              │◀── tool calls ──────────────│
 │◀ agent/run/token (streaming tokens) ───────│
 │◀ agent/run (final response) ───────────────│

Supported ACP methods

Method	Description
`agent/run`	Start a conversation turn (streaming via `agent/run/token` notifications)
`agent/list`	Return agent metadata (name, description, capabilities)

Embed in your own binary

[dependencies]
edgecrab-acp = { path = "../edgecrab-acp" }

use edgecrab_acp::AcpServer;
use edgecrab_core::Agent;

let agent = Agent::default_builder()?.build()?;
AcpServer::new(agent).run_stdio().await?;