JamJet policy as an OpenAI Agents SDK guardrail, enforcing policies on agent actions.
The portable policy layer for AI agent actions.
One policy file. One audit trail. Across hooks, guardrails, MCP gateways, SDKs, and custom runtimes.
@jamjet/claude-code-hook — JamJet policy as a Claude Code PreToolUse hook
@jamjet/openai-guardrail — JamJet policy as an OpenAI Agents SDK guardrail
@jamjet/mcp-shim — JamJet policy as an MCP stdio interceptor
@jamjet/cedar: Cedar policies (AgentCore, AGT style) validated against a published tool-call schema and composed with policy.yaml
@jamjet/cli — Unified audit show + approve across all adapters
All four adapters share a single policy.yaml format and a single audit JSONL schema.
See conformance/ for the spec these adapters all satisfy.
Evaluate the currently loaded JamJet policy against a candidate MCP tool name. Returns the matched rule, the decision (allow / block / require_approval / audit), and the matched glob pattern. Use this to dry-run a tool call before issuing it, or to explain why a previous call was blocked. Read-only and side-effect free — does not write to the audit log.
Inputs: tool_name (required), arguments.
List every rule in the currently loaded JamJet policy, in declaration order. Each entry includes the rule index, action (allow / block / require_approval / audit), and the glob pattern it matches. Use this to inspect the active policy without reading the YAML file directly, or to verify a rollout placed rules in the expected order. Read-only.
Inputs: no named parameters.
Report the path to the policy file currently loaded by this server, plus the number of rules it contains and the policy schema version. Use this once at session start to confirm the server has the expected policy attached. Read-only.
Inputs: no named parameters.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 3 tools
Each tool has a clearly distinct purpose: evaluate a specific tool against policy, list all rules, and report loaded policy info. No overlap between them.
All tools follow the consistent pattern 'policy_verb[_noun]' using snake_case, making them predictable and easy to understand.
With 3 tools, the server is tightly scoped to policy inspection and evaluation, fitting its purpose well without unnecessary bloat.
The set covers the essential operations for a read-only policy server: evaluating a candidate, inspecting rules, and confirming the loaded policy. No missing functionality for its intended use.
Real-time MCP & A2A prompt-injection, jailbreak, and tool-schema poisoning firewall for AI agents.
Security & DLP proxy for MCP: tool-poisoning scans, PII redaction on tool args/results. Beta.
MCP enforcement layer that intercepts AI agent actions and blocks rule violations before execution.
Scan prompts for injection attacks, redact PII, and audit LLM SDK usage from any MCP client
