What it does
The APC Security Reference Server is an MCP server for evaluating action security in synthetic scenarios. It is part of the open-core APC project, which binds an authorized decision to the exact action it approves. The project canonicalizes and hashes each proposed action (action, resource and parameters) at ingress, tracks how intent carries across agent-to-tool handshakes using signed witness tokens, and can sit in front of sinks such as Stripe, AWS IAM, Kubernetes admission controllers or databases to verify execution independently.
It is relevant when an LLM agent consumes untrusted data that could lead to prompt injection, or when tool-call parameters might change between authorization and execution. The project's local lab simulates attack classes such as mid-flight target value mutation, prompt injection tool hijacking and reservation race contention, so that defenses can be tested against them.
Testing runs against a local SQLite sandbox. The project is designed to be hosted by the customer, inside the customer's own execution boundary.
Safety
Our team has not reviewed this resource. Only automated rules checked it; resources with high-risk signals are not published. These are signals, not a guarantee that installing it is safe. Try it in a sandbox project first.
- Runs code or changes an assistant's behaviour
Sources
- Official MCP Registry first seen 13 Nov 2026 · last seen 13 Nov 2026 · its page on the source
APC Security Reference Server
Reference MCP server for synthetic APC action-security evaluation and local SQLite sandbox testin...
v0.1.0-alpha.1As shown on registry.modelcontextprotocol.io when we last read it