Agent Explorer

Quick Start

Get up and running with Agent Explorer in just a few steps.

1. Sign In

Open Agent Explorer and sign in with your email or Google account. New accounts are auto-seeded with one Example Agent and one Example MCP Server so you have something to play with right away.

After an update, a What’s new popup summarizes the latest changes the first time you open the app while signed in. Choose Got it, close it, or select Read the guide to dismiss it. Dismissal is remembered in this browser until a new announcement is published; clearing browser storage or using a different browser may show it again.

2. Land on the Explorer

After sign-in you'll land on the Explorer page. Workspaces organize your resources; the main list combines A2A agents and MCP servers.

  • In List view, use All, Agents, or MCP Servers to narrow the list.
  • Search across resource names and endpoints, filter by source, or change the sort order. Filters apply across all pages.
  • One paginator controls the results. On desktop, Split view shows both resource types in separate columns and hides the All/Agents/MCP Servers filters. Search, source filtering and sorting still apply. Returning to List view restores the previous type filter; switching views starts at page 1. On narrow screens, the list and its type filters remain available. List view is the default.
  • Search, filters, ordering, and page are kept in the URL, including when you return from a resource. The list/split preference is saved in this browser.
  • On smaller screens, open Workspaces to select a workspace.
  • Workspace names stay on one line in the sidebar. Hover over a name or focus it with the keyboard to see the full name.
  • On desktop, drag the divider beside the workspace sidebar to resize it. Your layout is remembered in this browser. Double-click the divider to reset, or focus it and use the left/right arrow keys to adjust its width.

If the example resources from sign-up are still there, click the Example Agent to skip ahead to step 4.

3. Connect an Endpoint

Click Connect endpoint in Explorer. Paste an HTTP(S) endpoint URL and leave Auto detect selected, or choose A2A agent or MCP server explicitly.

  1. Click Check connection. Auto detection tries AgentCard discovery first, then an MCP initialize handshake when no valid card is found.
  2. Review separate results for reachability, discovery, authentication, protocol version and advertised capabilities. Discovery does not send agent messages or invoke MCP tools.
  3. Review the connection name and owner. The selected workspace and its owner are inherited; changing the owner clears the workspace assignment.
  4. Click Connect and open Test to save the connection.

Authentication can prevent detection. Select A2A or MCP, enter a name and use Connect without verification, then configure credentials on the resource. Checks run from the Explorer server without credentials. An accessible AgentCard does not prove that its message endpoint works; an MCP handshake does not prove that every tool can be invoked. Version and capability reports are discovery evidence, not compliance results.

For a known first test, expand Need an endpoint to try? and choose Try the built-in echo agent. Connect, then send Hello; the expected reply is Echo: Hello.

Use Create or import → Import AgentCard for a JSON file, Create mock agent for an A2A mock, or Create mock MCP server for a server with configurable capabilities. Each action opens its dedicated form; use Cancel or close the dialog to choose a different action. An imported AgentCard can be connected to a real endpoint or cloned with Mock this agent.

4. Open the Agent and Test It

Selecting an agent in the Explorer opens its detail page with tab navigation:

  • Agent Card — view and validate the AgentCard
  • Configure — name, URL, headers, ownership, and (for mocks) skills, response behaviour, API keys
  • Test — send A2A messages and watch task state in real time
  • Flows (mocks only) — visual node-based response logic, per skill
  • Webhooks (mocks only) — push-notification subscriptions and delivery log
  • Integrate — copy-pasteable code snippets for your language of choice
  • Export tests — preview or download Postman and curl tests

Real agents land you on Test by default. Newly created mocks land you on Configure so you can fill in skills and flows before sending traffic.

The Test toolbar contains New conversation, History, Inspector, and Request settings. History and Inspector share one optional side panel on desktop and open as a dialog on smaller screens. Switching panels preserves your draft. Resource information starts collapsed. Supports streaming describes the agent capability; task status shows current activity.

Example messages, above the composer, lists every example declared by each skill, including imported AgentCards. Collapse or reopen the panel at any time; it remains available after sending messages and when reopening an existing conversation. Selecting an example previews it; Use example fills an empty composer without sending or overwriting your draft.

To test:

  1. Open the Test tab.
  2. Type a message, or preview a declared skill example and click Use example. Examples only fill an empty composer; an existing draft is preserved. Review the request, then press Send (or Enter).
  3. The agent's response and the live task state appear in the conversation; expand any message to see the raw JSON-RPC request and response.

5. Inspect Logs

The Logs link at the bottom of the agent's tab sidebar opens a full request/response history with timestamps, latency, and pretty-printed JSON.

Mock Agent Endpoint URLs

When you create a mock agent, it's immediately reachable at:

{base-url}/api/mock-agents/{id}/v1.0.0
{base-url}/api/mock-agents/{id}/v0.3.0

The protocol version is always explicit in the URL — pick the one matching the spec your client supports.

The mock's AgentCard discovery document lives at:

{base-url}/api/mock-agents/{id}/.well-known/agent-card.json

Mock MCP Server Endpoint URLs

Mock MCP servers expose a single JSON-RPC endpoint plus a discovery document:

{base-url}/api/mock-mcp-servers/{id}/mcp
{base-url}/api/mock-mcp-servers/{id}/.well-known/mcp-server.json

Built-in Test Agent

For the absolute quickest test with zero setup, point any client at the built-in echo agent:

{base-url}/api/mock-agent

It echoes messages, accepts special commands like delay 3 and status failed, and supports both synchronous and streaming responses.

What's a Workspace?

Workspaces are optional. They're folders that group related agents and MCP servers so you can hand a single bundle to a teammate or scope work for a project. You can use Agent Explorer without ever creating one — pick Create or import → Workspace later if you want the organization.

To organize any existing agent or MCP server, whether connected or mocked, open its options menu in Explorer, choose Move To, then select a workspace. Resources can belong to more than one workspace, so this adds the selected workspace without removing any existing workspace memberships. You can also add or remove memberships from Configure → Workspaces on agent and MCP server detail pages.