SORCERESS
Browse the wiki

Browser, MCP Servers & Skills

Updated July 4, 2026Open the tool

WizardGenie includes three power features that extend what the AI agent can do during a chat: a built-in Browser, configurable MCP Servers, and reusable Skills. Together, they let the agent inspect web pages, connect to external tool servers you control, and follow your preferred project rules or workflows.

Open WizardGenie here: https://sorceress.games/wizard-genie

What these tools do

Browser

The Browser panel gives WizardGenie an embedded browsing surface. In the desktop app, you can use it like a small browser, and the AI agent can also use it during chat to navigate, inspect pages, click controls, type into fields, scroll, use browser history, read visible text, and capture visual context when available.

In web mode, the live browser panel is disabled because an interactive embedded browser cannot be rendered inside the web version. The agent can still browse through a headless browser and report its findings in chat.

MCP Servers

MCP Servers let WizardGenie connect to external tool servers that expose additional capabilities to the agent. Once a server is added and enabled, WizardGenie can discover that server’s available tools and let the chat agent use them when relevant.

You can add local command-based servers or remote URL-based servers. The MCP server list is global configuration, shared across projects rather than tied to a single game.

Skills

Skills are user-defined instruction packs. A skill can contain reusable rules, conventions, knowledge, or workflow guidance that you want the agent to follow. Enabled skills are loaded into the agent on your next chat turn.

Skills are useful for things like:

  • Coding conventions for a specific engine or framework
  • Art direction rules for a project
  • Naming, file organization, or commit-message preferences
  • Repeatable workflows such as “always verify generated assets before continuing”
  • Team-specific constraints the agent should remember

Browser panel

Desktop and web behavior

WizardGenie’s Browser panel behaves differently depending on where you are running it:

  • Desktop app: the Browser panel renders an interactive live browser view with navigation controls and a page viewport.
  • Web mode: the panel shows a placeholder explaining that the visible browser is disabled. The AI agent can still browse the web in the background and stream results into chat.

If a chat request requires visible manual browsing, use the desktop app. If you only need the agent to research, summarize, or inspect pages, web mode can still work through chat.

Using the Browser manually

In the desktop version of WizardGenie, open the Browser panel to view the embedded browser.

The toolbar includes:

  • Back — returns to the previous page when available.
  • Forward — moves forward in browser history when available.
  • Reload — reloads the current page. The icon animates while a page is loading.
  • Address/search bar — enter a full web address, a bare domain, or a search query.
  • Security indicator — a lock appears for secure pages; a globe appears otherwise.
  • Search button — submits the text in the address/search bar.
  • Agent indicator — appears while the AI agent is actively browsing.
  • Loading bar — appears at the top of the page area while navigation is in progress.

To navigate manually:

  1. Open the Browser panel.
  2. Click the address/search bar.
  3. Enter one of the following:
    • A full web address
    • A domain name
    • A search phrase
  4. Press Enter or click the search icon.

If you enter a search phrase instead of a recognizable address, WizardGenie opens a web search for that phrase. If you enter a domain without a protocol, WizardGenie treats it as a secure web address.

Agent browsing behavior

When the agent uses the Browser, it can:

  • Navigate to a page or search query
  • Read the current page title and address
  • Inspect interactive page elements such as buttons, links, text boxes, checkboxes, radio buttons, tabs, sliders, menus, switches, selectors, and similar controls
  • Click elements
  • Type text into fields
  • Clear a field before typing when needed
  • Submit a field after typing when needed
  • Scroll up or down
  • Go back or forward
  • Read visible page text
  • Capture screenshots for visual context when available
  • Press common keyboard keys or shortcuts when needed

The agent uses a fresh page inspection to identify clickable and typable elements. If a page changes after loading, the agent may need to inspect it again before interacting with the correct control.

When the agent clicks or types in the desktop browser, WizardGenie may briefly highlight the target area. This helps you see what the agent is interacting with.

Good Browser prompts

Use direct prompts that tell the agent what you want it to inspect and what kind of output you need. For example:

  • “Use the browser to research current Phaser input handling guidance and summarize what matters for this project.”
  • “Open the documentation page I pasted, find the section about animations, and explain the recommended workflow.”
  • “Search for examples of cozy game UI layouts and suggest three menu structures.”
  • “Go to this page, inspect the form, and tell me what information it asks for before entering anything.”

For pages with dynamic content, ask the agent to re-check the page after scrolling or after a control expands.

MCP Servers panel

The MCP Servers panel manages external tool servers that WizardGenie can use from chat.

Key features

From the MCP Servers panel, you can:

  • Add new MCP servers manually
  • Import MCP server definitions from JSON
  • Enable or disable individual servers
  • Edit server name, transport, connection details, and credentials
  • Remove servers
  • Refresh the list when servers are added or changed elsewhere
  • Let the WizardGenie chat agent help manage MCP servers, then refresh the panel

The panel shows how many configured servers are currently enabled, such as 2/3 on.

Server list

When you open MCP Servers, the list view shows each configured server with:

  • An on/off toggle
  • The server name
  • A transport label
  • A short connection summary
  • A remove button

If no servers have been added, the panel displays No servers yet.

Use the toggle to enable or disable a server without deleting it. Click a server row to edit it. Click the trash icon to remove it.

The panel also refreshes when the window regains focus, which helps keep it in sync if servers were changed outside the panel. You can still click Refresh manually at any time.

Transport types

WizardGenie supports two MCP server connection styles.

stdio (local command)

Use this for a server that runs as a local command on your machine.

Settings:

  • Name — friendly name shown in WizardGenie. Required.
  • Transport — choose stdio (local command).
  • Command — the command used to start the server.
  • Arguments — optional command arguments, entered as a space-separated line.
  • Environment — optional environment values, one per line in KEY=value format.
  • Enabled — whether WizardGenie should make this server available to the agent.

http (remote URL)

Use this for a server exposed at a remote URL.

Settings:

  • Name — friendly name shown in WizardGenie. Required.
  • Transport — choose http (remote URL).
  • URL — the server URL.
  • Headers — optional request headers, one per line in KEY=value format.
  • Enabled — whether WizardGenie should make this server available to the agent.

Add an MCP server manually

  1. Open the MCP Servers panel.
  2. Click Add server.
  3. Enter a Name.
  4. Choose a transport:
    • stdio (local command) for a local command server
    • http (remote URL) for a remote server
  5. Fill in the transport-specific fields.
  6. Leave Enabled checked if you want the agent to use it.
  7. Click Save server.

WizardGenie saves the list and makes enabled servers available to the agent.

The Save server button is disabled until the server has a name. Use Cancel or the back arrow to leave the editor without saving.

Import MCP servers from JSON

If you already have MCP server configuration in JSON form:

  1. Open the MCP Servers panel.
  2. Click Import JSON.
  3. Paste the JSON into the editor.
  4. Click Import.

WizardGenie imports valid server definitions and enables imported servers by default. If an imported server has the same name as an existing server, the imported server replaces the existing one with that name.

If the JSON is invalid or contains no recognizable servers, the panel displays an error message. Use Cancel or the back arrow to return to the server list without importing.

Enable, disable, edit, refresh, or remove a server

From the server list:

  • Use the toggle switch to enable or disable a server.
  • Click a server row to edit its settings.
  • Click the trash icon to remove it.
  • Click Refresh to reload the list, especially after the chat agent or another process adds or changes MCP servers.

When changes are being saved, the footer briefly shows Saving…. Otherwise, the footer reminds you that the WizardGenie chat agent can also add and manage MCP servers if you ask it in chat.

How MCP tools appear to the agent

When a server is enabled and connects successfully, WizardGenie discovers the tools exposed by that server. The agent can then choose those tools during chat when they are relevant to your request.

If a server is disabled, disconnected, misconfigured, or unavailable, its tools are not available until the issue is resolved.

Skills panel

The Skills panel lets you create reusable instruction sets for the AI agent.

Key features

From the Skills panel, you can:

  • Add custom skills
  • Enable or disable skills independently
  • Edit skill names, usage hints, and instruction text
  • Remove skills
  • Save changes automatically after editing, toggling, or removing
  • Load enabled skills into the agent on the next chat turn

The panel shows how many skills are enabled, such as 3/5 on.

If no skills have been added, the panel displays: No skills yet. Add one to teach the agent reusable rules, conventions, or workflows.

Skill fields

Each skill has:

  • Name — required. A short label for the skill.
  • When to use — optional. A short hint describing when the skill applies.
  • Instructions — the main content the agent should follow. Markdown is fine.
  • Enabled — whether the skill should be included for upcoming chat turns.

In the skill list, each row shows the skill name and either its When to use hint, a preview of the instructions, or (no description) if neither is available.

Add a skill

  1. Open the Skills panel.
  2. Click Add skill.
  3. Enter a Name.
  4. Optionally fill in When to use with a short hint.
  5. Write the skill’s rules or knowledge in Instructions.
  6. Leave Enabled checked if you want it active.
  7. Click Save skill.

The Save skill button is disabled until the skill has a name. Use Cancel or the back arrow to leave the editor without saving.

Edit, enable, disable, or remove a skill

From the Skills list:

  • Use the toggle switch to enable or disable a skill.
  • Click a skill row to edit it.
  • Click the trash icon to remove it.

Changes save automatically. Disabled skills remain saved, but they are not loaded into the agent context.

Enabled skills load on your next chat turn after they are saved or toggled on.

Writing effective skills

Good skills are specific, concise, and action-oriented. The agent is more likely to follow a skill when it is clear about when and how to apply it.

Good examples:

  • “When writing Phaser scene code, keep scene names short and readable. Prefer composition over inheritance. Explain likely side effects before changing gameplay systems.”
  • “For pixel-art prompts, always describe palette, silhouette, resolution, animation state, and background transparency.”
  • “Before modifying save-related code, ask whether save data compatibility matters.”
  • “When proposing file organization, keep generated assets grouped by feature and use consistent lowercase names.”

Avoid turning every preference into a separate always-on skill. If a skill is only relevant sometimes, include a clear When to use hint or disable it until needed.

Give the agent research access

  1. Open the Browser panel if you are using the desktop app and want to watch or guide navigation.
  2. Ask the agent to research a topic, inspect a documentation page, or compare examples.
  3. Watch for the Agent indicator while it browses.
  4. If the page has dynamic content, ask the agent to re-check, scroll, or inspect again.

Example requests:

  • “Use the browser to inspect the latest Phaser documentation for input handling, then summarize what applies to this project.”
  • “Find examples of cozy game UI layouts and suggest a menu structure.”
  • “Open this documentation page and identify which steps are relevant to exporting a web build.”

Connect external tools with MCP

  1. Open MCP Servers.
  2. Add or import the server configuration.
  3. Make sure the server is enabled.
  4. Click Refresh if the server was added or changed outside the panel.
  5. Start a chat and ask the agent to use the connected tool.

If a server requires local software, credentials, or a running service, confirm those are set up before asking the agent to use it.

Teach project conventions with Skills

  1. Create a skill for your coding, art, or design conventions.
  2. Keep the instructions specific and concise.
  3. Enable it.
  4. Continue chatting; the skill applies on the next turn.

For example, you might create one skill for code style, one for art direction, and one for review workflow. Enable only the ones that matter for the current task.

Combine all three

For complex tasks, Browser, MCP Servers, and Skills can work together:

  1. Enable a project convention skill.
  2. Add any external tool server the agent needs.
  3. Ask the agent to research documentation in the Browser.
  4. Have it apply the skill’s rules while using the connected tools.
  5. Review the result and ask follow-up questions.

This is useful for tasks like integrating a library, comparing APIs, generating assets against a style guide, or checking documentation before making changes.

Tips & troubleshooting

Browser says the panel is disabled

You are likely using WizardGenie in web mode. The visible embedded browser is only available in the desktop app. The AI agent can still browse in the background and report results in chat.

The agent says the Browser is not connected

In the desktop app, open the Browser panel first. The agent needs the Browser panel to be available before it can control the visible browser.

The agent clicked the wrong thing

Pages can change after an inspection, especially dynamic sites. Ask the agent to inspect the page again before clicking, or manually navigate to the correct page in the Browser panel.

The agent cannot find an element

Try these steps:

  1. Wait for the page to finish loading.
  2. Ask the agent to inspect the page again.
  3. Scroll to reveal the control.
  4. If the page is highly visual or inaccessible, describe the target manually.

The agent cannot read enough of a long page

Ask it to read a narrower section, scroll, or focus on a specific part of the page. Very long pages may need to be handled in chunks.

An MCP server does not appear to work

Check that:

  • The server is enabled.
  • The command or URL is correct.
  • Required environment values or headers are present.
  • Any local server dependencies are installed.
  • Any required local service is running.
  • The server can start or respond outside WizardGenie.
  • You clicked Refresh after external changes.

Imported MCP JSON fails

Make sure the pasted text is valid JSON and contains recognizable server definitions. If import succeeds but a server does not connect, edit the imported server and verify its transport settings.

A skill is not affecting the agent

Enabled skills load on the next chat turn. If you just enabled or edited a skill, send a new message. Also check that the skill is toggled on and that its instructions are clear enough for the agent to apply.

Too many skills make the agent less focused

Keep skills concise and targeted. Disable skills you do not currently need, and use the When to use field to clarify when a skill applies.

FAQ

Are MCP Servers project-specific?

No. The MCP Servers panel manages global configuration shared across projects.

Do disabled MCP servers get removed?

No. Disabling a server keeps it saved but prevents WizardGenie from making that server’s tools available to the agent.

Do disabled skills get deleted?

No. Disabled skills remain in your list and can be re-enabled later.

Can the chat agent manage MCP servers?

Yes. The MCP Servers panel notes that the WizardGenie chat agent can add and manage MCP servers. If it does, click Refresh in the MCP Servers panel to reload the latest configuration.

Can the Browser fill out forms?

The agent can click fields, type text, optionally clear existing text, and submit by pressing Enter. Some complex pages, login flows, captchas, or inaccessible controls may still require manual help.

Can the Browser use keyboard shortcuts?

Yes, the agent can press common keys and shortcuts when needed, such as Enter, Escape, Tab, arrow keys, delete/backspace, and modifier-key combinations.

When do skills apply?

Enabled skills load on your next chat turn after they are saved or toggled on.

Can I use Markdown in skill instructions?

Yes. Markdown is fine for skill instructions. Keep the content readable, specific, and focused on what you want the agent to do.