This project is not covered by Drupal’s security advisory policy.
Edit your site via chat.
Use the same AI chat interface you've become familiar with to edit your entire site's content and configuration.
Content editors and site administrators can use the Site Agent to simplify their workflows and modify the website via LLM. Site Agent integrates with Drupal's existing permission sets and other AI modules to expose a simple chat window that allows people of any technical ability to manage even the most complex websites.
Features
- Integrates into the Drupal AI ecosystem
- Uses Drupal's existing permissions model
- Publishes nothing without approval
- Offers previews of all changes, both config and content
- Works out of the box with config-in-code
- All tool calls are logged with arguments
- Submodule automatically wires in AI Context
- Stage and deploy content with Content Deployment
Tested Providers
Site Agent should work with any AI provider, but these are the ones we have tested:
Submodules
- Site Agent MCP: offers an account's toolset to remote MCP clients, such as Claude, through MCP Server, with the same checks, approvals and audit log as the chat. With MCP Server OAuth, clients sign in with OAuth.
- Site Agent Tool Belt: Tool API tools for form and view display placement, menus, role permissions, block placement, recent log messages, and Views edits and clones.
- Site Agent Pathauto: Tool API tools to list and set Pathauto patterns.
- Site Agent AI Context: gives the model the site's AI Context items, in the system prompt and through a tool for the page being drafted. Installing it adds that tool to the toolsets that write content.
Requirements
- Drupal 11.3 or later, and PHP 8.3.
- AI 1.5 or later, with a provider module that returns streamed tool calls, such as Anthropic or OpenRouter.
- Tool API 1.0 beta.
- The shipped toolsets use Tool Belt's tools, or Content Deployment's where it is installed. Drafts need Content Moderation or revisionable content.
- Optional: AI Context 1.0.0-rc1 or later for Site Agent AI Context, and MCP Server 2.0.0-beta5 or later for Site Agent MCP.
Project status
Releases stay alpha or beta until Tool API reaches 1.0, since its APIs can still change between betas. The submodules are experimental for the same reason.
Similar projects
AI Agents runs agents on the AI module, but runs each tool as soon as the model picks it and keeps no record of what the tools did. Site Agent holds a call that needs approval until someone approves it, and records every call. It also has things AI Agents leaves to other modules or doesn't do: its own chat panel, a permission per toolset, before-and-after previews of each change, publishing kept for people, structure changes only where configuration is writable, content staged in workspaces, and the same checks for remote MCP clients. Its approval contract follows the AI Agents proposal #3586020; once that lands, Site Agent's runner, which sits behind an interface, can be rebuilt on AI Agents.
Project information
- Project categories: Artificial Intelligence (AI), Content editing experience, Site structure
- Ecosystem: AI (Artificial Intelligence)
- Created by tbcs on , updated
This project is not covered by the security advisory policy.
Use at your own risk! It may have publicly disclosed vulnerabilities.
Releases
Development version: 1.0.x-dev updated 8 Oct 2026 at 13:51 UTC
