13 tools. 4 can modify or destroy data without limits.
2 destructive tools with no built-in limits. Policy required.
Last updated:
Destructive tools (cancel_prediction, clear_image_cache) permanently delete resources. There is no undo. An agent calling these in a retry loop causes irreversible damage.
Execute tools (create_and_poll_prediction, create_prediction) trigger processes with side effects. Builds, notifications, workflows — all fired without throttling.
Intercept sits between your agent and Replicate. Every tool call checked against your policy before it executes — so your agent can do its job without breaking things.
npx -y @policylayer/intercept scan -- npx -y @mcp-replicate cancel_prediction:
rules:
- action: deny Destructive tools should never be available to autonomous agents without human approval.
get_collection:
rules:
- rate_limit: 60/minute Controls API costs and prevents retry loops from exhausting upstream rate limits.
Yes. The Replicate server exposes 2 destructive tools including cancel_prediction, clear_image_cache. These permanently remove resources with no undo. Intercept blocks destructive tools by default so they never reach the upstream server.
13 tools across 3 categories: Destructive, Execute, Read. 9 are read-only. 4 can modify, create, or delete data.
One line change. Instead of running the Replicate server directly, prefix it with Intercept: intercept -c replicate.yaml -- npx -y @@mcp-replicate. Download a pre-built policy from policylayer.com/policies/replicate and adjust the limits to match your use case.
Starter policies available for each. Same risk classification, same one-command setup.
Set budgets, approvals, and hard limits across MCP servers.
npx -y @policylayer/intercept init