Good list. One thing to add under keeping descriptions static: check it from the outside after deploy too. Save the tools/list response you reviewed, hash each tool's name, description and schema, and diff that against what production actually returns. A config change or a dependency bump can change what the model sees without anyone touching the reviewed release.
Also worth deciding on purpose whether tools/list itself needs auth. "No unauthenticated route should reach a tool" covers calls, but an open tools/list still publishes every description and schema to anyone with the URL.
X1Y1
AI features that work in production. RAG, agents, MCP, cloud and security for startups.
Good list. One thing to add under keeping descriptions static: check it from the outside after deploy too. Save the tools/list response you reviewed, hash each tool's name, description and schema, and diff that against what production actually returns. A config change or a dependency bump can change what the model sees without anyone touching the reviewed release.
Also worth deciding on purpose whether tools/list itself needs auth. "No unauthenticated route should reach a tool" covers calls, but an open tools/list still publishes every description and schema to anyone with the URL.