For general information about the Validio MCP Server, see Validio MCP Server. For installation and configuration, see MCP Server Setup.

Effective Prompting

• Be specific with source names and field names.
• Provide context about what you're trying to achieve.
• Ask for explanations to learn best practices.
• Iterate based on the assistant's responses.

Model Selection

• Choose a model that is capable of tool calling and reasoning to make sure it can handle more complex tasks. Some more complex tasks might require a frontier model (Claude Sonnet, Gemini Pro) and might not be fulfilled correctly by a smaller, faster model (Claude Haiku, Gemini Flash).
• Choose a model with enough context length if you want to fetch large amounts of data from Validio for analysis within the LLM (for example for listing many incidents).

Authentication

• The Validio MCP server will inherit the permissions and access of the user account you use to connect to it.
• Depending on your MCP client, the AI assistant may force you to re-authenticate with the Validio MCP at certain times. Make sure you are always authenticated to avoid errors when using the Validio MCP.

Security

• Destructive operations are always refused. The MCP server rejects any operation that deletes, removes, or resets a resource, including operations added to Validio in future releases. This applies regardless of the permissions held by the account you connect with, so no configuration or data can be destroyed through the MCP server.
• Non-destructive changes are still possible. Your AI assistant can create and update resources such as sources, validators, filters, windows, and segmentations, within the permissions of the account used to connect. Review what the assistant proposes before you confirm it.
• Scope the connecting account. The permissions of the user account you connect with are the main boundary on what your AI assistant can reach. Connect with an account whose role matches the access the assistant should have.
• To make the connection read-only, disable the write tools in your AI assistant: the general-purpose GraphQL tool (graphql-execute) and the resource-creation tools, such as create-validator, create-filter, create-window, create-segmentation, and manage-source. Disabling graphql-execute on its own does not make the connection read-only, because the creation tools work independently of it. The remaining tools continue to support read-only workflows such as incident analysis and validator recommendations.