What is SentinelMCP?
SentinelMCP is a gateway that sits between MCP clients and upstream MCP servers, checking requests before forwarding tool calls. It is aimed at teams operating MCP servers that need controls for identity, tool access, and usage budgets. The gateway can pause a call when a budget limit or usage spike is reached, giving a human a chance to approve or deny that specific request. Its core gateway is MIT-licensed, with additional production features available in Pro.
SentinelMCP Features:
Origin Validation: Checks the request origin on every call, allowing loopback by default and using an explicit allowlist for other origins.
Identity Verification: Verifies ID-JAG tokens against JWKS, checks issuer and audience claims, and binds the token to the gateway's resource.
Tool Policies: Runs a policy gate on each request and can deny tool calls before they reach an upstream MCP server.
Budget Controls: Estimates token use from JSON-RPC payload structure and enforces per-team budgets across a seven-day sliding window.
Spike Detection: Compares current usage with a team's recent baseline to flag sudden increases, independently of the hard budget limit.
Human Approval: Returns a request-bound resume token when a budget breach pauses a call, allowing a human to approve or deny it through the resume endpoint.
MCP Transports: Supports Streamable HTTP and legacy HTTP+SSE, with Streamable HTTP as the primary transport.
SentinelMCP Benefits:
Reduced Exposure: Requests that fail origin, identity, or policy checks stop before they can reach connected tools, narrowing the routes an unwanted call can take.
Earlier Cost Signals: Budget and spike checks can surface unusual usage while an agent is running, rather than leaving teams to discover the increase only in a later bill.
Safer Exceptions: A human can make a one-call decision about an over-budget request without turning off identity and policy checks for the resumed call.
Bounded Budget State: The fixed-memory budget ledger avoids unbounded growth from keeping a separate request log for each call.
Deployment Flexibility: The request and response path is stateless, so teams can place it behind a load balancer while accounting for in-memory budget and SSE state.
Use Cases:
Production MCP Gateway: Place SentinelMCP in front of an MCP server that gives agents access to databases, payment APIs, or internal services.
Agent Cost Control: Set per-team usage limits for agent workflows that may repeatedly call expensive tools or enter a retry loop.
Sensitive Tool Access: Check caller identity and tool policy before forwarding requests to services that should not be available to every MCP client.
Budget Exception Review: Route an over-budget call to a human for a decision when a workflow has a legitimate reason to proceed.
Multi-Client Migration: Support MCP clients using Streamable HTTP while retaining legacy HTTP+SSE for clients that have not migrated.

