The MCP server of Foundry is great, but I would like to avoid any LLM deleting things by mistake.
Bonus: Is it possible to turn off deletion/gate some particular actions ?
As a note, from https://www.palantir.com/docs/foundry/palantir-mcp/security#write-access in the meantime: All tools that can perform write actions are either non-destructive or require a human to approve the changes.
Not available yet but we are working on a new feature named Ontology-MCP which will let you convert any Dev Console application to an MCP server. this will include Platform API you enabled on that Dev Console application.
With that you should be able to control with APIs you are enabling.
Update for posterity: This feature (Ontology MCP) now exists and it serves the OP’s needs using Application Restrictions.
My recommendation is to create a separate Developer Console application for your MCP server from any user-facing SDK app, so you can create independent restrictions. You will likely want to provide additional guardrails for agents connecting via MCP, compared to users in your front-end.