Talk to A2A Agents from the Terminal

proto is a first-class terminal client for A2A agents. Register an agent once, then proto agent <name> to open a streaming chat with it — no URL to retype, no browser. This is the client direction: proto reaches out and talks to a remote agent.

It's the mirror image of driving proto as an ACP coding agent, and together they make proto and the protoLabs protoAgent fleet talk both ways:

Direction	Protocol	proto's role	How
protoAgent → proto	ACP	agent/server	proto --acp; protoAgent's delegates plugin spawns proto as a coding agent
proto → protoAgent	A2A	client	proto agent <name>; proto streams a turn to the agent over A2A

protoAgent linkage

protoAgent is an A2A-native LangGraph agent runtime (and a fork-me template for a fleet of protoLabs agents). Every protoAgent serves an A2A 1.0 endpoint at POST /a2a plus a public agent card at /.well-known/agent-card.json. proto speaks exactly that surface, so any running protoAgent — or its forks (e.g. roxy) — is reachable from your terminal as soon as it's up.

Discovery is built in and mirrors protoAgent's own fleet broadcast/scan: proto probes localhost for agent cards, so a protoAgent running on :7870 shows up in proto agents with no registration at all.

A2A compatibility

proto implements the A2A 1.0 client surface that a2a-sdk (≥ 1.1) serves:

• Transport — JSON-RPC 2.0 over POST /a2a, with the mandatory A2A-Version: 1.0 header.
• Methods — SendStreamingMessage (SSE), SendMessage, GetTask, CancelTask.
• Streaming — Server-Sent Events whose frames are the A2A oneof (task / statusUpdate / artifactUpdate / message), rendered live.
• Conversation — a per-session contextId pins one memory thread, so the agent remembers across turns.
• Discovery — the agent card (skills, capabilities, declared extensions, auth schemes).
• Extensions — renders the protoLabs A2A extensions: cost-v1 (token usage), confidence-v1, and tool-call-v1 (live tool activity).
• Auth — Authorization: Bearer … (from a configured env var) or X-API-Key; open endpoints (common locally) need no credentials.

Because it targets the protocol, not the implementation, proto works against any conformant A2A 1.0 agent, not just protoAgent.

Use it

# See every agent — registered and auto-discovered on your machine — with status
proto agents

# Register a named shortcut (so you never type the URL again)
proto agent add roxy https://roxy.internal:7870

# Open a streaming chat, addressed by name
proto agent roxy

proto agent <name> resolves the name from your registry first, then falls back to live discovery — so you can connect to a just-discovered agent by name even before you register it.

Registering with auth

# Send a Bearer token held in an env var on every request
proto agent add roxy https://roxy.internal:7870 --bearer-env ROXY_TOKEN

# Or a static header / an X-API-Key env var
proto agent add roxy https://roxy.internal:7870 -H "Authorization: Bearer …"
proto agent add roxy https://roxy.internal:7870 --api-key-env ROXY_API_KEY

Registered agents live in the a2aAgents section of ~/.proto/settings.json (user scope) or .proto/settings.json (project scope, via --scope project).

Commands

Command	What it does
proto agents	List registered + discovered agents with up/down status
proto agent <name>	Connect and chat (resolves via registry, then discovery)
proto agent add <name> <url>	Register an agent under a name
proto agent remove <name>	Unregister an agent
proto agent list [--no-scan]	Same as proto agents (--no-scan skips discovery)

See also

• Drive proto as an ACP Coding Agent — the other direction: proto as the agent an orchestrator drives
• Run Headless — proto -p and stream-json
• Configure Models & Auth — endpoints and credentials