Policy enforcement for MCP pipelines
Add guardrails for agent tool calling. Enforce access control, rate limits, and cost management across MCP pipelines without code changes. Swap one URL and you're protected.
How it works: deterministic, in-path policy enforcement → View the policy model
The Problem
MCP servers expose powerful tools with zero restrictions
GitHub, Slack, databases - these integrations give AI agents direct access to production systems with no access controls, no rate limits, no audit trail.
One misconfigured agent can delete repositories, spam customers, or rack up unexpected API costs. You need guardrails before going to production.
How It Works
Add protection with a single URL change
Before
Direct connection, no protection
mcp.github.com/v1After
Protected with policies
abc123.keypost.ai/mcpCreate a Keypost
Point it to your upstream MCP server
Add policies
Access control, rate limits, DLP, and more
Swap the URL
Update your agent config and you're protected
Comprehensive Policy Enforcement
Protect your systems with flexible, composable policies
Access Control
Allow or deny specific tools with granular permissions
- • Allowlist dangerous operations
- • Role-based access rules
- • Tool-level permissions
Parameter Constraints
Validate and restrict parameter values
- • Regex pattern matching
- • Value ranges and blocklists
- • Type validation
Time-based Restrictions
Schedule tool availability by day and time
- • Business hours enforcement
- • Prevent weekend deployments
- • Timezone-aware rules
DLP & PII Scanning
Block sensitive data patterns automatically
- • SSN, credit card detection
- • API key scanning
- • Custom pattern blocking
Rate Limiting
Control request frequency at multiple levels
- • Per-minute/hour/day quotas
- • User & team-level limits
- • Recipient-based throttling
Cost Management
Track and limit spending by team or project
- • Budget tracking & alerts
- • Response caching with TTL
- • Cost attribution tags
Built for Performance
Rust-powered policy enforcement designed for speed and reliability
Typical policy evaluation adds minimal latency
Single instance throughput for high-scale deployments
End-to-end response time at the 99th percentile
SSE passthrough for real-time responses
Change rules instantly without restarting or dropping requests
Works seamlessly with Claude Desktop, Cursor, and any MCP-compatible agent
Secure policy namespaces with team-level separation
Built on Rust and Tokio. Performance targets based on architecture design.