it is now possible to use an AI agent via MCP server directly within Radzen Studio. That is great but what about security? Which data is shared with the agent and how is security-related information (e.g. passwords from connection strings in appsettings.json) handled? Does the agent get full access to this data or is this hidden form the agent?
Also, does the agent get full access to the entire project/solution including the configured database connections?
The MCP server built into Radzen Blazor Studio runs on your machine and only listens on localhost. It also rejects requests from browser pages, so a website you visit can’t call it. Radzen doesn’t send your project or your data anywhere through it.
Through MCP, the agent can do what the tools allow: list the projects, pages, entities and database connections, edit pages and themes, and add data sources.
Connection strings:list_connections never returns passwords. The agent sees only the connection name, database type, server and database name. When it needs a stored connection, such as to list tables or scaffold, it passes the name, and Studio reads the credentials itself.
Database access: the tools only read the schema (tables, views, stored procedures) to scaffold code. There’s no tool that queries your data.
Note that MCP isn’t the only way an agent reaches your project. Agents like Claude Code, Copilot or Cursor already have their own access to your files, so they can open appsettings.json with or without Radzen’s MCP server. What an agent can read on its own depends on that agent’s permission settings.