Extensions has three kinds of information: switchable modules, JSON definitions, and read-only tools registered by other plugins through MazeGraph. The entries are loaded from the current site, so installed plugins and module availability can change what you see.
Modules
Optional modules can be turned off independently. Required modules are marked Always on and cannot be switched off here. A disabled module stops providing its feature; it does not uninstall a plugin. If a module fails during startup, MazeGraph reports the failure and leaves that module off for the affected request.
| Built-in module | Can be switched off? | What it controls |
|---|---|---|
| Activity | Yes | Activity history, undo, and background jobs. |
| Content Studio | Yes | Brief-based content drafting and its Studio workspace. |
| Helpers | Yes | Delegation to the read-only Site researcher. The Ask a helper tool must also be enabled. |
| MCP server | Yes | MazeGraph’s own MCP host. External sharing must also be configured in Settings → MCP. |
| Notes | Yes | Recall notes and save them under the active session mode and policy. |
| Site tools | Yes | MazeGraph’s read-only content, media, and site tools. |
| Usage | Yes | Token and usage summaries. |
| WooCommerce tools | Yes | Read-only catalog and store summaries plus optional undoable product and order actions; requires WooCommerce. |
| Chat workspace | No | Chat sessions, tools, approvals, and undo. |
| Definitions | No | The catalogue of agents, templates, starters, and policy definitions. |
| Draft tools | No | Undoable draft and related content writes used in chat and Content Studio; assistant calls follow session mode and policy. |
| Tools and extensions | No | The Tools, Agents, and Extensions administration screen. |
Plugins or themes may register additional modules, which appear in the live list with their source. The current free MazeGraph source does not include built-in modules named Performance Tools or Nodes; if either appears on a site, it is supplied by an installed extension.
Definitions
Definitions are structured JSON, not executable code. They describe the site’s agent roles, Content Studio templates, starter prompts, and policy rules.
| Definition type | What it describes |
|---|---|
| Agent | An agent’s role, instructions, tool patterns, and optional step limit. |
| Studio template | A content task’s fields and the stages it uses. |
| Starter prompt | A suggested question or request in the workspace. |
| Policy rule | A rule that can restrict an action or require approval. |
Definitions may come from MazeGraph, a plugin, the active theme’s maze-graph.json, or this site. Choose JSON to inspect a definition. Only editable, site-owned definitions can be deleted here; a deletion can be undone from Activity. A short About this site section may also show lines from definitions that MazeGraph adds to each session’s instructions.
Tools added through MazeGraph
Other plugins can register read-only tools through MazeGraph’s Registrar. They appear in this section and in the Tools catalogue, where an administrator can review and enable them. The plugin that registers a tool remains responsible for its behavior; MazeGraph still applies its own availability and policy checks.
The WordPress MCP Adapter can host MazeGraph’s MCP server or serve WordPress Abilities through its own default server. MazeGraph can use the official adapter as the host or use its own server, according to MCP settings. Only eligible read-effect tools can be exposed through MCP, and the MazeGraph-managed server checks policy and records tool calls. The separate Adapter-sharing switch uses the direct Ability path; those abilities still run their permission callback and MazeGraph’s direct-call policy guard.
For code-based extensions, tool registration, and JSON definition schemas, see the developer reference.