Safe AI Tool Framework
ToolRampart
toolrampart
An open-source Python framework for exposing normal Python functions as safe AI-agent tools and MCP-compatible tools.
FastAPI for safe AI tools, built for teams that want agent capability without unrestricted production access.
ToolRampart sits between agents and production functions so useful capabilities can be exposed with guardrails. It validates inputs with Pydantic, checks actor scopes, runs policy functions, supports approvals for risky calls, deduplicates retries with idempotency keys, enforces rate limits, applies timeouts and retries, and emits audit logs with optional OpenTelemetry signals.
MCP-compatible tools
Pydantic validation
Approval workflows
What it handles