Agent Policy Enforcement Mcp
Define per-agent-pair IAM rules so orchestrators can only call downstream agents (e.g. billing) under explicit conditions before you ship multi-agent stacks.
Overview
Agent Policy Enforcement MCP is a MCP server for the Ship phase that enforces per-agent-pair IAM policies for agent-to-agent calls.
What is this MCP server?
- Per-agent-pair IAM for agent-to-agent (A2A) calls with declarative policy strings
- Streamable HTTP remote at api.meok.ai plus stdio PyPI package (v1.0.4)
- Python runtime with MCP 2025-12-11 server schema
- Orchestrator-to-specialist constraints (example: billing only when amount is below a threshold)
- GitHub source: CSOAI-ORG/agent-policy-enforcement-mcp
- Server version 1.0.4
- 1 streamable-http remote endpoint
- 1 PyPI stdio package (Python runtimeHint)
What problem does it solve?
Multi-agent setups default to overly broad tool access, so one compromised or mis-prompted orchestrator can invoke sensitive peers without guardrails.
Who is it for?
Indie builders shipping orchestrator plus specialist agent graphs who need lightweight IAM without building a custom auth service.
Skip if: Single-agent chat with one MCP and no A2A delegation, or teams that already enforce all access through a central API gateway with identical semantics.
What do I get? / Deliverables
After you register this server, each A2A edge can be allowed or denied by explicit policy so only qualified agent pairs run high-risk tools.
- Registered MCP connection for policy evaluation on A2A edges
- Declarative per-pair policies aligned to your orchestration graph
- Runtime gate before sensitive cross-agent tool invocations
Recommended MCP Servers
Journey fit
Policy gates belong on the canonical Ship → Security shelf because they are the authorization layer you enable before agent fleets handle real user or payment flows. Subphase security matches express policies like amount caps and caller/callee pairs rather than generic backend wiring.
How it compares
MCP policy enforcement for agent pairs, not an in-repo Claude skill for writing application code.
Common Questions / FAQ
Who is agent-policy-enforcement-mcp for?
Solo and small-team builders running multiple cooperating agents who need declarative rules on which agent may call which other agent or tool.
When should I use agent-policy-enforcement-mcp?
Use it when you introduce orchestrator-to-specialist A2A flows—especially around billing, PII, or privileged tools—before or as you go to production.
How do I add agent-policy-enforcement-mcp to my agent?
Add the PyPI stdio package agent-policy-enforcement-mcp to your MCP client config, or point a streamable-http client at https://api.meok.ai/v1/a2a/policy-enforcement per the server manifest.