A2A / MCP Protocols

Connect an A2A agent

Once your A2A server agent is deployed and reachable, connect it to a Copilot in Agent Studio → A2A Agents. The Copilot then discovers your agent’s skills and delegates matching tasks to it automatically.

Before you begin

Your remote A2A server agent must run over HTTPS and be reachable.
It must serve an Agent Card at /.well-known/agent.json (or a custom URL).
You need the agent’s base URL and authentication credentials.
A Copilot can connect up to 20 A2A agents.

The agent list

Agent Studio → A2A Agents lists every agent connected to this Copilot:

Column	What it shows
Name	The display name you gave the agent. This is also how it appears in the Actions filter in Conversations.
Base URL	The root URL of the remote agent.
Auth	The auth type in use — None, Bearer Token, or API Key.
Agent Card	The name from the fetched Agent Card and its skill count. A green check means the card was fetched and validated.
Status	Toggle an agent on or off. A disabled agent stays configured, but the Copilot won’t invoke it.

Use Add Agent to connect a new one, or the ⋯ menu on a row for per-agent options.

Connection details

Path: Agent Studio → A2A Agents → Add Agent

Field	Required	Description
Display Name	Yes	A friendly name for this agent (e.g. “Status Monitor Agent”). Shown in the Copilot’s tool list.
Base URL	Yes	The root URL where the remote A2A agent is hosted (e.g. `https://my-agent.example.com`).
Agent Card URL	No	Override the default card location. Leave blank to use `<base-url>/.well-known/agent.json`.
Protocol Version	Yes	Currently A2A v0.3 (auto-selected).

Authentication

Auth type	When to use	Fields
None	Agent is open or network-restricted (no auth headers sent)	—
Bearer Token	Agent expects `Authorization: Bearer <token>`	Token value
API Key	Agent expects `X-API-Key: <key>`	Key value

After saving, click Fetch Agent Card to validate the connection and pull in your agent’s skills, then choose which skills to expose to the Copilot.

Advanced settings (optional)

Forward user context

Toggle on to pass the current user’s identity to the remote agent via the X-Foldspace-User-Id header. You can also include a structured user-context data part:

Source field	Default key	Description
User Email	`user_email`	The authenticated user’s email address
User ID	`user_id`	The user’s unique identifier
Subscription ID	`subscription_id`	The customer’s subscription identifier

Additional headers

Add static custom headers (for API versioning, routing, etc.) sent on every request to your agent.

Connection timeouts

Setting	Default	Range	Description
Connect Timeout	10s	1–300s	Max time to establish a connection
Request Timeout	120s	1–600s	Max time to wait for a single HTTP response
Task Timeout	300s	1–900s	Max total time for an A2A task to complete

Troubleshooting

Start in Agent Studio → Conversations and open the Actions filter — your A2A agent is listed there by its display name. Select it and click Apply to narrow the list to conversations that delegated to it. Open one, click View Analysis on the agent message, and check the Actions tab: each A2A call shows a Remote Agent badge, its status, and — when expanded — the Arguments sent, the Response returned, and the execution time. That’s usually enough to tell a timeout apart from an auth failure or a bad payload.

Problem	Solution
”Failed to fetch agent card”	Verify the Base URL is reachable, the Agent Card URL serves valid JSON, and your auth credentials are correct.
Agent shows but the Copilot doesn’t invoke it	Check that the agent is Enabled and has at least one Exposed skill.
Authentication errors at runtime	Confirm the token or API key hasn’t expired or been revoked.
Agent times out during execution	Increase the Task Timeout in Advanced settings (default 300s).

A2A server agents — concepts and the Agent Card.
A2A quickstart — build your first agent.
A2A advanced patterns — LangGraph, TypeScript, Java, native.
A2A agents (Agent Studio) — the product overview.