1.3.0

WPGPT MCP Bridge v1.3.0

Overview

WPGPT MCP Bridge v1.3.0 builds on the standalone MCP and secure sandbox work from v1.2.0. The main goal of this release is to make the plugin easier to use with ChatGPT and other MCP clients when many abilities are enabled, while keeping the detailed internal security model intact.

Instead of exposing every detailed query, inspect and apply ability separately, WPGPT now exposes compact grouped abilities such as wpgpt/posts, wpgpt/media, wpgpt/settings and wpgpt/danger. This reduces discovery noise and helps avoid excessive MCP requests on hosting environments that rate-limit aggressively.

Highlights since v1.2.0

• Added compact MCP ability discovery.
• ChatGPT now sees one grouped ability per module instead of many detailed abilities.
• Added action routing through action=query, action=inspect and action=apply.
• Kept detailed abilities as the real internal security boundary.
• Added compact wpgpt/danger facade for filesystem, sandbox and PHP actions.
• Hardened compact execution so every action revalidates:
  • detailed ability allowlist;
  • WordPress user permissions;
  • Safe mode / read-only state;
  • filesystem read/write/delete permissions;
  • explicit deletion confirmation settings.
• Improved the settings page with a compact discovery preview.
• Compact preview now separates actions that are available from actions blocked by current permissions.
• Aligned detailed ability category names with compact module names.
• Moved the compact discovery preview into a collapsed-by-default section to reduce visual clutter.
• Renamed confusing counters from “active” to “selected” where appropriate.
• Improved delete exposure policy so deletion requires both the filesystem delete permission and explicit deletion confirmation.

Compact abilities

Examples of compact abilities exposed to MCP clients:

wpgpt/posts
wpgpt/drafts
wpgpt/terms
wpgpt/media
wpgpt/plugins
wpgpt/settings
wpgpt/users
wpgpt/database
wpgpt/danger

A compact ability accepts an action and parameters:

{
  "action": "query",
  "parameters": {}
}

Internally, WPGPT routes the request to the corresponding detailed ability, for example:

wpgpt/posts + action=query   -> wpgpt/posts-query
wpgpt/posts + action=inspect -> wpgpt/posts-inspect
wpgpt/posts + action=apply   -> wpgpt/posts-apply

Danger module

The wpgpt/danger compact ability groups advanced filesystem, sandbox and PHP actions.

Possible internal actions include:

list_directory
read_file
write_file
edit_file
delete_file
execute_php
disable_file
enable_file

These actions remain permission-controlled. For example, write/edit actions require filesystem write permission, deletion requires both delete permission and explicit deletion confirmation, and PHP execution remains a high-risk capability that should only be enabled for controlled maintenance.

Security notes

The compact discovery layer does not bypass security. It is a presentation and routing layer. The detailed allowlist and permission checks remain authoritative.

Dangerous operations should remain disabled unless needed. PHP execution is powerful and runs in WordPress context, not in a true OS-level sandbox.

Upgrade notes

1. Upload and activate the plugin normally.
2. Open WPGPT MCP in the WordPress admin sidebar.
3. Review the MCP user.
4. Review global permissions.
5. Review selected detailed abilities.
6. Open the compact discovery preview to confirm what ChatGPT will see.
7. Reuse the existing MCP token; normal updates and ability changes do not require regenerating it.

Compatibility

• WordPress 6.6+
• PHP 8.1+
• MCP Adapter is not required for the ChatGPT connection.

1.2.0

WPGPT MCP Bridge v1.2.0

Overview

WPGPT v1.2.0 is the first consolidated release of the standalone WPGPT MCP Bridge experience.

This version no longer requires MCP Adapter for the ChatGPT connection. WPGPT now exposes its own native WordPress MCP endpoint, adds a safer ability allowlist model, and introduces the new secure sandbox workflow for advanced filesystem and PHP-assisted maintenance tasks.

Highlights

• Native standalone MCP endpoint for ChatGPT and compatible clients.
• MCP Adapter is no longer required for the ChatGPT flow.
• Persistent MCP token across plugin updates and ability changes.
• Dedicated WordPress admin sidebar menu.
• Improved settings layout:
  • permissions first;
  • ChatGPT/MCP access second;
  • full-width ability list;
  • database controls below abilities.
• New Danger category for advanced abilities.
• Secure sandbox at wp-content/wpgpt-sandbox/.
• Optional persistent sandbox PHP loader.
• Safe mode with .crashed marker after fatal sandbox errors.
• .disabled support for enabling/disabling sandbox PHP files.
• Backups before dangerous file operations.
• JSONL audit logs for dangerous actions.
• Clearer permissions UI with Safe mode: block changes.

New Danger abilities

wpgpt/danger-list-directory
wpgpt/danger-read-file
wpgpt/danger-write-file
wpgpt/danger-edit-file
wpgpt/danger-delete-file
wpgpt/danger-execute-php
wpgpt/danger-disable-file
wpgpt/danger-enable-file

These abilities are intentionally powerful. Keep them disabled unless needed for a controlled task.

Security improvements

• Safer permissions model for read, write, edit, delete, and destructive operations.
• Read-only/safe mode blocks incompatible dangerous permissions.
• Critical path protections for filesystem operations.
• PHP write/edit flows restricted to the sandbox.
• Optional sandbox loader is disabled by default.
• Crash detection prevents broken sandbox PHP from repeatedly taking the site down.
• Dangerous actions are logged.
• Existing files are backed up before overwrite/edit/delete flows.

Token behavior

Updating the plugin does not require regenerating the MCP token.

Changing allowed abilities or permissions does not require regenerating the MCP token either. Save the settings and the next discovery request will reflect the current configuration.

Use Regenerate token only when you intentionally want to invalidate the previous connection.

Upgrade notes

1. Upload and activate the plugin normally.
2. Open WPGPT MCP in the WordPress admin sidebar.
3. Review the selected MCP user.
4. Review permissions and exposed abilities.
5. Keep dangerous abilities disabled unless needed.
6. Verify the ChatGPT endpoint still works with the existing token.

Recommended production posture

• Enable Safe mode: block changes when not actively working.
• Keep deletion disabled by default.
• Keep wpgpt/danger-execute-php disabled by default.
• Use a dedicated MCP user.
• Keep only the abilities needed for the current workflow enabled.
• Review backups and audit logs after dangerous operations.

1.1.0

Modular WordPress MCP bridge with generalized query / inspect / apply abilities, safer execution, structured outputs, and scalable AI-oriented architecture.

1.0.3

feat(audit): add safe options, DB orphan, media usage and discussion audit abilities".

Add new read-only audit capabilities for options, autoload, transients, orphaned DB relations, media usage, unused media, uploads auditing, SEO analysis snapshot and discussion overrides.

Also validate and keep contains-based search behavior for postmeta and JetEngine option scanning.

1.0.2

fix(search): normalize metadata filters to contains matching

Update postmeta-search to use partial LIKE matching.

Keep backward compatibility with previous params and align JetEngine option scan with contains-based filtering.

1.0.1

Fix: corrected the MCP endpoint shown in the admin for ChatGPT and MCP clients.
The generated Bearer token URL now points to the MCP Adapter server endpoint:

/wp-json/mcp/mcp-adapter-default-server?wpgpt_token=...

instead of the internal plugin route:

/wp-json/wpgpt-mcp/mcp?wpgpt_token=...

This fixes connection issues caused by the admin displaying the wrong MCP URL.

v1.0.0

Revertir versión a 1.0.0