Skip to content

TOOL_JUSTIFICATIONS.md

Latest commit

 

History

History
28 lines (17 loc) · 1.39 KB

TOOL_JUSTIFICATIONS.md

File metadata and controls

28 lines (17 loc) · 1.39 KB
Raw

Tool justifications for OpenAI app approval

AgentStack GPT Actions expose two operations only:

Operation OpenAPI Purpose
list_actions GET /mcp/actions Read-only discovery of MCP action names and caps
execute_tool POST /mcp Execute one registered action by { "tool", "params" }

Copy the justification blocks below into OpenAI review forms.

list_actions

Annotations: readOnlyHint=true, openWorldHint=false, destructiveHint=false, idempotentHint=true

Justification:
Lists available AgentStack MCP actions from our backend catalog. Read-only; no mutations. Calls only https://agentstack.tech/mcp/actions. Same request returns the same catalog snapshot for a given registry version, so it is idempotent. Models should call this before unfamiliar mutations.

execute_tool

Annotations: readOnlyHint=false, openWorldHint=false, destructiveHint=true, idempotentHint=false

Justification:
Executes a single registered AgentStack MCP action by name with parameters. Mutations (create project, payments, deletes) depend on the chosen action; the GPT must confirm destructive or money-moving operations with the user. Calls only our MCP API at https://agentstack.tech/mcp. Not globally idempotent — each call may create new resources. Discovery via list_actions keeps the action surface explicit and reviewable.