// BRING YOUR MCP SERVER
Your agents already have tools. Now govern them.
Point any MCP client at the MakeYourMCP gateway and every tool call is policy-checked, injection-scanned, and written to a tenant-scoped audit log — before it reaches your systems.
Three ways ungoverned agents hurt you
Prompt injection
A poisoned document can instruct your agent to call delete_table. Without a gateway, nothing stands between that instruction and execution.
No RBAC
MCP clients trust every tool a server exposes. Interns and production agents get identical power.
No audit trail
When something goes wrong, "which agent called what, when, and why was it allowed" has no answer.
One URL between your agents and your tools
Register your MCP server's URL
Write an allow/deny policy — read_* allowed, delete_* denied
Point your MCP client at your gateway endpoint. Watch decisions land in the audit feed.
audit feed
DENIED delete_table · policy: read-only-analyst · injection: none · 12ms
Where MakeYourMCP sits
| MakeYourMCP | Cerbos | LLM gateways (e.g. Bifrost) | Manual allow-lists | |
|---|---|---|---|---|
| Built for | MCP tool-call security | General-purpose authorization policy engine | Routing and observability for LLM API traffic | Per-client config files |
| MCP-native | Yes — point any MCP client at the gateway | Integration is yours to build | Governs model calls, not MCP tool calls | Yes, but per machine |
| Audit trail | Per-call, tenant-scoped, with decision + reason | Depends on your integration | Request-level logs | None |
| Injection detection | Built in, per call | Not its job (authorization only) | Varies by product | None |
| Time to first policy | Minutes — URL + API key | A policy-as-code project | An infra deployment | Editing every client by hand |
// FAQ
Bring-wedge questions
Do I have to change my MCP server?
No. The gateway proxies your existing server. You register its URL, attach a policy, and point clients at the gateway endpoint instead.
What does a policy look like?
Allow/deny rules over tool names with glob patterns — allow read_*, deny delete_* — with deny winning and a default action when nothing matches. Role/resource constraints are on the roadmap.
Where do denied calls go?
Nowhere — that's the point. The call is blocked at the gateway, the agent gets a structured denial, and the attempt is recorded in your audit log with the matched rule.
Can I try it without a credit card?
Yes. The free plan includes one server, one agent, and a seven-day audit trail. No card until you upgrade.
// GO_DEEPER
Go deeper on securing MCP
The Complete Guide to Securing the Model Context Protocol (MCP)
The definitive technical guide on securing Model Context Protocol integrations, preventing prompt injections, shielding credentials, and auditing agent tool execution.
Understanding Role-Based Access Control (RBAC) in MCP
How to design, write, and enforce granular authorization perimeters for autonomous AI agents using role-based access controls.
Isolating Credentials in AI Workflows via HSM Vaults
Prevent prompt injections from exposing access keys. Discover the server-side credential hydration pattern for Model Context Protocol (MCP).