diff --git a/custom/knowledge/mcp/bc-mcp-connects-to-curabis-production-not-customer-sandbox.md b/custom/knowledge/mcp/bc-mcp-connects-to-curabis-production-not-customer-sandbox.md
new file mode 100644
index 0000000..c030e8f
--- /dev/null
+++ b/custom/knowledge/mcp/bc-mcp-connects-to-curabis-production-not-customer-sandbox.md
@@ -0,0 +1,66 @@
+---
+bc-version: [all]
+domain: mcp
+keywords: [businesscentral, mcp, production, sandbox, bc-mcp-bridge, psa, invoicing]
+technologies: [al, mcp]
+countries: [w1]
+application-area: [all]
+---
+
+# BC MCP Connects to CURABIS Production, Never Confuse It With a Customer Sandbox
+
+## Description
+
+The `businesscentral` MCP server (`bc-mcp-bridge.js`) connects to CURABIS's
+own Project Management 365 Business Central environment — the system
+CURABIS runs its actual time registration, invoicing, and project tracking
+through. It is production data, full stop, on every project, for every
+developer.
+
+This is a categorically different system from the **customer's** BC
+sandbox/test environments that AL development and automated domain tests
+run against. Both may be open in the same session, both may be casually
+called "BC," but they are not the same system and must never be treated as
+interchangeable in stakes or disposability.
+
+## Rule
+
+An agent must never describe, treat, or retry a `businesscentral` MCP call
+as if it were hitting disposable test data. Reads remain safe to perform
+unprompted (see `bc-mcp-scope-tasks-to-repository` and sibling rules) — this
+rule is about STAKES, not permission: don't casually re-fire a slow or
+failed BC MCP call without first naming that it targets production, and
+never suggest "let's just try it again" the way one might for a sandbox
+call.
+
+## Why
+
+CURABIS runs its own invoicing and time registration through this system.
+An agent that treats it as "just another read-only API, safe to poll" has
+collapsed two different properties into one: "safe to read" (true) and
+"low-stakes / disposable" (false). Conflating them leads to careless
+behavior — repeated polling, casual retries, treating slowness as a
+shrug-worthy glitch — against a system real invoices depend on.
+
+## Anti-Pattern
+
+    Developer: "That BC MCP call is slow."
+    Agent: "Want me to just retry it?"
+
+No acknowledgment that "BC" here means CURABIS's own production PM365
+tenant, not the customer sandbox the rest of the session's AL work is
+otherwise happening against.
+
+## Compliant
+
+    Developer: "That BC MCP call is slow."
+    Agent: "That's CURABIS's own production PM365 environment — I won't
+    just retry it casually. Want me to try once more, or hold off until we
+    know why it was slow?"
+
+## Scope
+
+Applies to every CURABIS project with the `businesscentral` MCP server
+configured. The target environment is identical across all of them
+(CURABIS's own PM365 tenant), regardless of which customer's codebase the
+session is otherwise working in.