A2A Server

Estimated reading: 8 minutes

The A2A Server turns any agent workflow in RobilityFlow into a shared, reusable service. Other workflows can call it from within the same project and across environments.

It exposes the whole agentic workflow through standard HTTP endpoints. Other AI agents, workflows and client applications can then use the open Agent2Agent (A2A) agent component to find it, send it tasks and get results back.

How it works

A2A Server: The published workflow that performs the work and offers its skills to others.
A2A Agent component: The component added to other workflows to send tasks to the A2A Server.

Key capabilities

a. Build once, reuse everywhere: A specialized agent is maintained in one workflow and used within the same project and across environments without duplicating its setup.
b. Automatic Agent Card and endpoint: Publishing generates the Agent Card and a secure connection endpoint.
c. Built-in testing: Agent responses can be tested directly in the Try It console.
d. Access control: Access to each published version is restricted using Robility Manager API keys.

Example: A dedicated Research Agent workflow searches company documents and summarizes data. With the A2A Server enabled, the Customer Support, Sales, and HR workflows can each query it on demand, and any update to the agent applies to all three.

Note: Support for external third-party A2A agents is coming soon.

When to Use A2A Server

Choose the A2A Server when a specialized Agent flow needs to be shared and invoked across multiple workflows without rebuilding its internal logic.

Requirement Suggested Approach
Share a specialized agent across multiple workflows in the same project A2A Server + A2A Agent component
Run a single self-contained agent inside one workflow Standard Agent or Deep Agent component
Invoke an external third-party REST API API Request component

Prerequisites

i. A2A functionality enabled on the Robility Flow server.
ii. A workflow containing Chat Input, an AI Model (or Agent), and Chat Output.
iii. Access to Robility Manager to generate a version-specific API key.

How It Fits into Robility Flow

1. Configuration and Publishing: A workflow creator builds an agent flow with Chat Input and Chat Output, enables Serve as an A2A Agent, configures its Agent Card metadata, and publishes the workflow.
2. Discovery: A consumer workflow or client inspects the Agent Card URL (/.well-known/agent-card.json) to verify the agent’s capabilities, version, and authentication scheme.
3. Request Invocation: The consumer sends a JSON-RPC message/send payload to the agent’s /jsonrpc endpoint with a valid x-api-key header (or invokes it directly via the A2A Agent component).
4. Workflow Execution: Robility Flow passes the incoming message into the published workflow’s Chat Input, executes the AI Model and any connected tools, and returns the final output from Chat Output as the JSON-RPC response.

How to Use A2A Server

1. Build the Agent Flow

Create a workflow containing the required A2A interface components:

a. Chat Input: Receives incoming A2A user messages.
b. AI Model (or Agent): Processes the incoming message and orchestrates tool calls.
c. Chat Output: Returns the final response to the caller.
d. Supporting Components: Add any tools, knowledge bases, or logic required for the agent’s task.

2. Enable and Configure the A2A Server

a. Open the Agent feature page from the right-side panel of the workflow canvas.
b. Enable Serve as an A2A Agent (if disabled, verify that Chat Input and Chat Output are present, or refresh the page).
c. Configure the Agent Card metadata

Field Description
Name* Specifies the display name of the A2A Agent.
Version Specifies the semantic or release version of the Agent.
Description Summarizes the Agent's purpose and supported tasks.
Tags Adds keywords to categorize the Agent for discovery.
Examples Provides sample prompts illustrating tasks the Agent can handle.

d. Save the configuration. The A2A Flow Status changes to Draft, and endpoint details appear in the Address.

3. Publish the Agent Flow

a. Publish the configured workflow.
b. Confirm that A2A Flow Status changes to Live.
c. Copy the generated Agent Card URL or cURL Command from the Address section.

Note: Republish the workflow whenever you modify its components or A2A configuration to apply the updates.

4. Generate an API Key

a. Open Robility Manager and navigate to Projects > [Your Project] > Workflows.
b. Select the published Agent workflow and click Generate Key.
c. Select the target workflow Version, set an expiry period in days, and click Generate.
d. Open Key History, click View, and copy the API key.

Note: API keys are bound to a specific published workflow version. Generate a new API key whenever you publish a new version.

5. Test the A2A Server

a. Open the Try It section on the Agent feature page.
b. Paste the generated API key into the X-API-Key field.
c. Enter a prompt in the Message field and select Send.
d. Verify that the agent processes the request and displays the expected response.

6. Connect from Another Workflow

a. Open the calling workflow in the same Robility Flow project and add the A2A Agent component.
b. Set the connection type to Internal and select the published Agent flow from the list.
c. Map the input message and connect the output to downstream components.
d. Run the workflow to send requests and receive responses from the A2A Server.

7. Disable the A2A Server

To stop serving the workflow over A2A, open the Agent feature page, toggle Serve as an A2A Agent off, and save the configuration. The A2A Flow Status reverts to Off, and the endpoints stop accepting requests.

Generated Endpoints and Authentication

Once saved, the Address section exposes two endpoints:

Endpoint Format Purpose
Agent Card URL https://flow.robility.ai/api/v1/a2a/<agent-id>/.well-known/agent-card.json Publishes the Agent's metadata, version, advertised skills, and authentication requirements for discovery.
JSON-RPC Endpoint https://flow.robility.ai/api/v1/a2a/<agent-id>/jsonrpc Accepts authenticated HTTP POST requests using the message/send method and returns agent responses.

Sample cURL Request

Select cURL in the Address section to copy a pre-formatted request:

curl -X POST 'https://flow.robility.ai/api/v1/a2a/<agent-id>/jsonrpc' \
  -H 'Content-Type: application/json' \
  -H 'x-api-key: <YOUR_API_KEY>' \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "message/send",
    "params": {
      "message": {
        "role": "user",
        "parts": [{ "kind": "text", "text": "Hello" }],
        "messageId": "1"
      }
    }
  }'

Replace <agent-id> with the published Agent ID and <YOUR_API_KEY> with the version-specific key from Robility Manager.

A2A Flow Status Reference

Status Meaning
Unavailable The workflow lacks required components (Chat Input or Chat Output). A2A serving cannot be enabled until they are added.
Off Serve as an A2A Agent is disabled.
Draft A2A serving is enabled and saved, but the workflow has not been published.
Live The workflow is published and actively accepting authenticated A2A requests.

Usage Example: IT Helpdesk Escalation

A primary Employee Support Workflow handles general workplace questions, while a specialized IT Diagnostics Workflow queries system logs and ticket databases:

1. The IT Diagnostics Workflow is configured with Chat Input, an AI Model connected to IT diagnostic tools, and Chat Output, then published as an A2A Server (Version 1.0).

2. When an employee asks the Employee Support Workflow to troubleshoot a VPN outage, the orchestrator routes the query to the A2A Agent component configured for Internal communication with IT Diagnostics Workflow.

3. The IT Diagnostics Workflow runs its checks and returns the diagnostic summary to the support workflow, which formats the final reply for the user.

Limitations and Guardrails

a. Same-project scope: Within Robility Flow, the A2A Agent component only supports communicating with published A2A Servers inside the same project. Cross-project internal calls are not supported.
b. Outbound external agents: While external Agent Card metadata can be fetched and previewed using an Agent Card URL, sending outbound requests from Robility Flow to external third-party A2A servers is not currently supported.
c. Mandatory interface components: A workflow must contain both Chat Input and Chat Output components to enable A2A serving.
d. Version-bound API keys: An API key generated for one published workflow version does not authenticate requests after a new version is published unless a new key is generated for that version.

Troubleshooting

Issue Possible Cause and Resolution
Serve as an A2A Agent toggle is disabled A2A Flow Status is Unavailable because Chat Input or Chat Output is missing. Add both components and refresh the page.
Agent Card URL returns no data The workflow is still in Draft or Off status. Save the A2A configuration, publish the workflow, and verify the URL.
A2A Flow Status remains Draft The configuration was saved, but the workflow was not published. Publish the workflow to transition the status to Live.
API key authentication fails The x-api-key header is missing, expired, or generated for an older workflow version. Generate a key for the active published version in Robility Manager.
Internal A2A communication fails The target Agent workflow is unpublished or belongs to a different Robility Flow project. Ensure both workflows reside in the same project and the target is Live.
Share this Doc

A2A Server

Or copy link

CONTENTS
Robility Chatbot
Robility Assistant
Online