polygraphso/Polygraph MCP gate

Fail the build if an MCP dependency grades D/F under the open polygraph behavioral litmus.

Trust Signals

name	description	required	default
servers	Explicit MCP refs to check (newline- or comma-separated). Merged with auto-discovery.	no	""
skills	Explicit skill directories to grade (newline- or comma-separated). Merged with auto-discovery.	no	""
discover	Auto-discover targets from MCP config files (.mcp.json, .vscode/mcp.json, .cursor/mcp.json). OFF by default — opt in only on trusted repos, since discovered targets are PR-controllable and grading runs their code.	no	false
min-grade	Minimum acceptable grade (A\|B\|C\|D). Default gates on D/F.	no	""
strict	Treat un-gradeable dependencies as failures instead of warnings.	no	false
working-directory	Directory to scan for MCP config files.	no	.
version	@polygraphso/litmus version to run.	no	0.36.0
api-url	Override the published-grade lookup API base URL. HTTPS is enforced (http only for localhost). Point only at the official endpoint or a mirror you trust — an attacker-controlled endpoint can return fabricated grades.	no	""
bearer	Bearer token sent as an Authorization header to a gated remote (https) target. Only for an explicitly trusted, pinned remote — never with discovery or on untrusted PRs; keep it scoped and short-lived.	no	""

name	description
result	pass or fail
failed	Number of dependencies that tripped the gate
report	JSON array of per-target results