Hi HN, I built Keypost after running into the same problem repeatedly with MCP servers. They expose powerful tools, but there is no standard way to control who can call what, with which parameters, or under what limits.
Keypost sits in the request path between an agent and an MCP server and enforces deterministic policies before tools run. That includes tool allow or deny rules, parameter constraints, rate and cost limits, and auditing. There are no SDKs or agent changes required. You swap one URL and policies apply consistently.
I wrote up how the policy engine works here:
https://keypost.ai/policy-model
We are in private beta and mainly looking for feedback from people actively experimenting with MCP servers or agent workflows.