This is powered by the Model Context Protocol (MCP), an open standard that lets AI applications connect to external tools and data sources. Because MCP is an open standard rather than a proprietary integration, the mWater MCP server works with a growing ecosystem of AI tools including Claude Desktop, Claude Code, OpenAI Codex, Cursor, and many others, and will keep working with new tools as they adopt the standard.
Giving an AI access to monitoring data is a serious step, especially for organizations working with sensitive information about communities, health facilities, and infrastructure. We designed the mWater MCP server around a simple principle: the AI works for you, within your boundaries, and you stay in control.
Concretely, that means:
Permission-aware access. The MCP connection inherits your mWater permissions and nothing more. The AI can only see the tables, sites, and records that your own account can see. It cannot reach data belonging to other users or organizations.
Read-only by default. All queries are strictly read-only. Queries time out at 60 seconds and are capped at a fixed row count and response size, so the connection is suited to questions and analysis, not bulk extraction. Very large exports are not supported via MCP and should continue to go through the Portal's standard export tools.
Human approval for every change. The AI never writes to your data. It can only prepare proposed changes, which appear on a review page in the portal where you can inspect each one and approve or reject it. No approval, no change. There are no exceptions to this rule.
Password-grade token security. Your MCP connection link contains a secret token and should be treated exactly like a password: never share it. The token is shown only once, at the moment you generate it. After that, mWater stores only an encrypted hash and neither you nor we can ever see the token again. If a token is ever exposed, simply generate a new link, which invalidates the old one.
One important thing to understand before you connect: data retrieved through MCP is processed by your AI assistant under that provider's terms of service, and depending on your settings, it may be used to train future AI models.
Before connecting, review your AI tool's privacy and data-use settings, and disable model training on your data if appropriate, especially if you work with sensitive or confidential information. Most major AI providers offer this control: look for settings named along the lines of "improve the model for everyone," "model training," or "data controls" in your AI tool, or use a commercial/enterprise plan, which typically excludes your data from training by default. Your organization's data protection policies may also have requirements here, so check with your administrator if you are unsure.
mWater itself does not use your data to train AI models, and the MCP server adds no new data sharing on our side, it simply gives your chosen AI tool the same view of mWater that you already have.
Outside of the MCP server, if you use the AI tools we have included in the platform such as the Integrator, AI translations, or dashboard widget creation, the data is never stored or used by our AI providers for training as per enterprise terms of service.
That's it. Open a conversation and try something like "What tables can you see in my mWater account?" or "How do I create a new survey in mWater?"
The MCP server is currently able to do the following:
Is my mWater data used to train AI models? Not by mWater. However, your AI provider may use conversation data, including data the assistant retrieves from mWater, for training, depending on your plan and settings. Check your AI tool's data controls before connecting, and disable training on your data if your work requires it.
What can the AI actually change in my data? Nothing on its own. Queries are read-only, and any addition, update, or deletion the AI prepares is only a proposal. Changes are applied solely when you review and approve them on the dedicated page in the mWater portal.
What does "permission-aware" mean? The AI connection has exactly the same data access as your mWater account. If you can't see a table or a site, neither can the AI. Administrators can therefore reason about MCP access the same way they already reason about user permissions.
Which AI tools work with this? Any tool that supports the Model Context Protocol with remote connectors, including Claude Desktop, Claude Code, OpenAI Codex, Cursor, and others. The ecosystem is growing quickly.
Can I revoke access? Yes. Generate a new MCP link from your account page (or remove the connector), and the previous token stops working. Treat tokens like passwords and rotate them if you suspect exposure.
Can I use this to export large datasets? No. Queries are capped in rows, response size, and execution time (60 seconds) by design. For large exports, use the portal's standard export features.
Does this cost anything? The mWater MCP server is part of your existing mWater account. Note that the AI assistant itself (Claude, Cursor, etc.) is a separate product with its own pricing.
What if the AI gives a wrong answer? AI assistants can make mistakes, which is exactly why the system is read-only with human-approved changes. Treat AI-generated analysis as a fast first draft, verify important findings in the portal, and review every proposed change carefully before approving.
This is a powerful new way to work with water and sanitation data, and we are just getting started. We would love to hear what works well, what's confusing, and what's missing. Share your feedback through info@mwater.co, and help us shape where this goes next.