On this page
ReGenWest speaks the Model Context Protocol. Any modern AI assistant can connect to the platform, read the live neighbourhood data, and even file an intervention for review, exactly as a citizen would through the submission form. This page explains what an assistant can do, how to connect one, and why the door is safe to leave open.
What an Assistant Can Do
The platform exposes five abilities to connected assistants. Four of them read, one of them writes:
- Platform statistics: the live totals behind the dashboards, from published actions and installed solar capacity to trees and nature-based surface.
- Intervention areas: the eleven areas with their sizes, sites, allowed categories and per-area indicator totals.
- Actions: the published interventions with their indicator contributions, filterable by category, area, actor and date.
- Categories: the thirteen intervention types with their descriptions and action counts.
- Submit an action: file a new intervention. A submission created this way is never published directly; it enters the same review queue as every entry made through the website, and it only appears on the map and in the indicators after the project team approves it.
Everything the read abilities return is information the website already shows to every visitor. An assistant sees the same neighbourhood you do.
In the Browser: WebMCP
The monitoring application carries a built-in WebMCP layer. Open the platform in a browser with a WebMCP-capable client, such as the WebMCP extension for Google Chrome, and the assistant can work with the page you are looking at through six tools: get_stats, get_categories, get_areas, search_actions, get_action_details and navigate_to_tab.
The scope is deliberately narrow. These tools read and navigate what the page has already loaded for you; they hold no credentials, change no data, and stop existing when you close the tab. Pairing happens through the small widget in the corner of the application, and nothing is installed on the server.

Connect a Remote Assistant
Assistants that support remote MCP servers connect to the platform directly over the internet:
https://YOUR-PLATFORM-DOMAIN/wp-json/mcp/mcp-adapter-default-server
The connection is authenticated. You need a ReGenWest account and an application password, which you create from your account profile. An application password is a separate, revocable key: it is not your real password, you can create one per assistant, and you can delete any of them at any moment without touching your account. The assistant then authenticates with HTTP Basic credentials formed from your username and that application password.
Claude Desktop and Gemini CLI
Both use a small bridge for remote servers. Add this to the client's MCP configuration and replace the placeholder with the Base64 encoding of username:application-password:
{
"mcpServers": {
"regenwest": {
"command": "npx",
"args": ["-y", "mcp-remote",
"https://YOUR-PLATFORM-DOMAIN/wp-json/mcp/mcp-adapter-default-server",
"--header", "Authorization:Basic ${AUTH}"],
"env": { "AUTH": "BASE64_OF_USERNAME_COLON_APP_PASSWORD" }
}
}
}
Clients with Native Remote Support
Clients that connect to remote HTTP servers directly, such as Gemini Antigravity, need only the server address and the same authorisation header:
{
"mcpServers": {
"regenwest": {
"serverUrl": "https://YOUR-PLATFORM-DOMAIN/wp-json/mcp/mcp-adapter-default-server",
"headers": { "Authorization": "Basic BASE64_OF_USERNAME_COLON_APP_PASSWORD" }
}
}
}
ChatGPT
ChatGPT connects through its connector settings for remote MCP servers. Add the endpoint above as a custom connector and supply the same credentials when prompted.



Reading the Raw Data API
Everything the platform shows is also served by the standard WordPress REST API, and assistants sometimes reach for it directly instead of the abilities above. If yours does, tell it to request only the fields it needs: the area collection carries the full map geometry of every polygon, which multiplies the payload more than a hundredfold and can overflow an assistant's reading window, a classic recipe for invented numbers. A filtered request stays small and complete:
https://YOUR-PLATFORM-DOMAIN/wp-json/wp/v2/area?per_page=20&_fields=id,title,area_size_m2,site_count,implementing_actors,allowed_project_categories
The MCP abilities avoid the problem by design: they return the compact records an assistant needs and never the geometry.
Security by Design
The AI door is open because it was built to be safe to open:
- Read access mirrors the public site. The read abilities return only what the dashboards and the public interface already display. Connecting an assistant reveals nothing new.
- Writing always means moderation. A submission from an assistant is created in a pending state, invisible to the public and absent from every indicator until the project team reviews and approves it. This rule is enforced on the server and cannot be bypassed by any client.
- Credentials are revocable and scoped. Application passwords are independent keys per assistant. Revoking one disconnects that assistant and nothing else. Your account password is never shared with any client.
- Transport and infrastructure. All traffic runs over TLS, and the server applies the same validation, rate protections and monitoring to assistant requests as to any other request.
Fair Use
The interface exists so that researchers, municipal teams and curious residents can put the neighbourhood's data to work. Use it freely for reading and analysis; keep submission for genuine interventions. The full technical reference for the interface lives in the platform documentation.
