Agent Explorer

Build the agents your orchestration needs—even before those agents exist.

Create mock A2A agents and MCP servers, give them predictable responses, and run them locally from VS Code. Test your orchestration offline without signing in or starting a separate backend.

What you can do

• Mock A2A and MCP endpoints. Test A2A 0.3 and 1.0 messages, or define MCP tools, resources, and prompts.
• Choose how each mock responds. Use fixed replies, echo, conditional rules, or simulated errors. Configure shared responses and overrides for individual skills or capabilities.
• Build visual response flows. Add branches, variables, payloads, delays, and errors in the flow builder.
• Route requests to different skills. Select one skill or use ordered regular-expression rules to choose a flow from the incoming message.
• Keep local URLs predictable. Run all mocks on one configurable port, with a separate path for each mock.
• Save mocks with your project. JSON definitions live in your workspace and can be version-controlled. Valid editor changes save automatically.
• Publish when you need hosted endpoints. Optionally publish to Agent Explorer, download cloud mocks, and update existing hosted copies while preserving their IDs.

Get started

1. Install Agent Explorer, published by agent-explorer, from VS Code’s Extensions view.
2. Open a project folder and select Agent Explorer in the activity bar.
3. Select Create Mock, choose A2A agent or MCP server, and give it a name.
4. Configure a response in Responses. For more complex behavior, select Edit in Flow Builder.
5. Select Start Local Server, then copy the mock’s endpoint into your orchestration.

The installed extension bundles its runtime. Local mocks need no separate Node.js installation, database, cloud account, or internet connection.

Define responses and routing

For A2A mocks, Route incoming messages determines which skill receives a request. Manual always chooses one skill; Pattern checks regex rules in order, with the first match winning.

Below the routing controls, Edit response for selects the response you are editing. Changing this selector does not change runtime routing.

Conditional replies check rules from top to bottom. Combine ALL or ANY conditions, set a reply for each rule, and provide an Otherwise fallback. Agent rules include Try a message for an offline preview. Use Run test to exercise the running mock, including routing and simulations.

Configure pattern-based routing and choose which response to edit in the Responses tab.

Design flows visually

Use the flow builder to connect response steps, branch on input, set variables, introduce delays, and simulate failures. Each skill or MCP capability can have its own flow, with a shared fallback for targets without an override.

A shared echo response in the Flow Builder: Start → Set Payload → End.

Set a fixed port

Select the Settings gear in the Agent Explorer sidebar, or run Agent Explorer: Settings from the Command Palette.

Set Port to a fixed value, such as 4310, for URLs starting with http://127.0.0.1:4310. Use the Workspace tab for this project, or User for your default across projects. Workspace settings take precedence.

• 0, the default, selects an available port automatically.
• A fixed port persists across server restarts and VS Code sessions.
• After changing the setting, stop and start the server to apply it.
• If a fixed port is occupied, the extension reports an error. It only uses a different port if you explicitly choose that option.

All enabled mocks share the port:

Keep mocks with your project

Definitions are stored in .agent-explorer/mocks/<id>.mock.json inside your project folder. In a workspace with multiple folders, the extension asks where to create each mock. Reopening a folder brings back its definitions; starting the server remains an explicit action.

Valid changes save automatically. The header shows Saving… and Saved after VS Code confirms the write. Cmd/Ctrl+S saves pending edits immediately. Invalid input remains editable; a running mock keeps its last valid definition until you correct and save the file.

Editor tabs show mock names. Renaming a mock preserves its ID, filename, and endpoint path. To edit JSON directly, use Reopen Editor With → Text Editor.

Optional cloud publishing

Run Agent Explorer: Connect Cloud Account and enter a management key from Agent Explorer under Settings → MCP access. Grant View mocks, Create mocks, and Edit mocks. The cloud URL is configured automatically, and the key is stored in VS Code SecretStorage.

Publish Mock creates a hosted copy with the local mock’s ID. Later publications update that same copy. Download Cloud Mock, Compare With Cloud, and Pull Cloud Changes let you bring hosted definitions into your workspace and review differences. Revision conflicts stop publication rather than overwriting newer cloud edits.

Publishing is explicit. Local definitions are not automatically synchronized to the cloud. New hosted mock endpoints are public by default; publishing asks for confirmation. Disconnecting leaves your local mocks available.

Requirements and current limits

• Desktop VS Code 1.95 or later. Browser-only vscode.dev is not supported.
• Workspace Trust is required to start local servers or publish definitions.
• The server binds to 127.0.0.1 and stops when its owning VS Code window closes or reloads. Saved definitions remain; runtime tasks and request logs are temporary.
• In a remote workspace, the server runs on the extension host’s machine. Automatic port forwarding has not been verified.
• Local LLM execution, push-notification delivery, public tunnels, and always-on hosting are not included.
• Cloud definitions that cannot run locally are rejected with compatibility details. There is no automatic merge or force overwrite.

Documentation

Extension guide · Cloud management access · Agent Explorer

Support

Ask a question or report a reproducible problem through Agent Explorer Community. Include your extension version, VS Code version, operating system, and the steps that reproduce the issue. Remove credentials and private payloads from logs or example mock definitions before sharing them.

License

Copyright © 2026 Nico Balestra. Agent Explorer is proprietary software, licensed under the terms included in LICENSE.txt. Third-party components remain subject to their respective licenses, included in the extension’s third-party notices.