Hi Cursor community,
I built a small MCP server called Decision Assistant.
It adds deterministic guardrails to AI coding workflows.
Instead of reviewing AI-generated code after the fact,
it interrupts risky actions before execution.
Example signals:
- files_touched
- diff_lines_total
- ship_gap_days
If a change looks risky the server returns:
ALLOW
REQUIRE_CONFIRM
BLOCK
In REQUIRE_CONFIRM mode the server issues a receipt
bound to a plan_hash. The user must explicitly re-run
the action with that receipt.
Repeated EXECUTE calls are idempotent.
The goal is not to replace AI coding tools,
but to add a safety layer around execution.
GitHub:
npm:
npx [email protected]
Would love feedback from people experimenting with MCP tooling.