AI & MCP Gateway
One gateway for models and tools
Route every LLM request and MCP tool call through one policy engine. Use Ferentin’s managed gateway, or run a Private Service Edge in your own cloud so the traffic stays inside your network.
What it covers
What it covers
One gateway for the two kinds of traffic agents send: model requests and tool calls. Each provider, server and agent is registered once and governed the same way.
LLM providers
OpenAI, Anthropic, Amazon Bedrock, Azure, Google Vertex AI, Mistral, xAI and more behind one OpenAI-compatible API. Ferentin handles translation, streaming, token tracking and cost accounting.
MCP servers
Remote servers from the catalog, by URL or from a server card. People connect as themselves in User Mode, and workloads connect as the agent in Agent Mode.
AI agents
Claude, Cursor, ChatGPT and custom agents register as clients with their own identity. People sign in through your identity provider, and agents can sign with their own key.
Model access per provider
Each provider instance exposes only the models you choose, by allowlist or blocklist. Separate instances keep dev and prod keys, teams or regions apart.
Routing and key rotation
Instances join the routing pool with a flag and tags. Add a new key as a second instance, watch traffic move, then retire the old one.
Spend visibility
Import usage and cost from a provider’s admin API key to reconcile metered figures with what the provider actually billed.
How it deploys
How it deploys
The control plane is always Ferentin’s. The gateway runs where you need it: managed by Ferentin as a Public Service Edge or as a Private Service Edge in your VPC. Both take the same policy.
- 1
Register providers and servers
Add each LLM provider instance with its credentials, test the connection and pick the models it exposes. Add MCP servers from the catalog or by URL the same way. Provider credentials are encrypted under your tenant’s own keys.
- 2
Choose where the gateway runs
Start on the managed gateway with nothing to deploy. When data must stay in your network, create an Edge Site and enroll a Private Service Edge, a single container with LLM Proxy, MCP Proxy or both. LLM and MCP traffic through a Private Service Edge never transits the Ferentin cloud.
- 3
Activate policy at the edge
Changes take effect on the managed gateway as soon as you save them. Private Service Edges enforce a signed policy bundle, so an activation publishes the pending changes and each edge picks them up within about five minutes.
What it enforces
What it enforces
Every request is checked before it reaches a model or a tool, and recorded after. The same checks run on the managed gateway and on every Private Service Edge.
Identity on every request
Each person signs in through your identity provider and each agent has its own identity, so every request names who made it and through which agent.
Model and tool policy
LLM policies decide which models each user, group or client may use. MCP policies grant servers and tools, deny them or hold them for an approver.
Usage limits
Token and dollar ceilings per minute and per day, per user, agent, LLM policy or provider instance, so a runaway agent does not become a runaway invoice.
Data protection
Detect PII, financial and health identifiers and secrets in prompts, responses and tool calls, then redact, block or log them in flight.
Prompt injection defenses
Content sanitization marks tool results as data, not instructions. MCP Guard scores results for prompt injection and tool poisoning and can block them.
Audit you control
Every request is recorded with its decision. Payload logging decides what is stored, and Zero Data Retention keeps no payloads at all.
Works with your stack
Works with your stack
The model providers behind the gateway. Several instances of one provider can run side by side, each with its own credentials and models.
An AI gateway is a single access layer between your users, agents and applications and the AI services they call. Ferentin’s gateway covers both LLM requests and MCP tool calls. It authenticates who is calling, checks the request against policy, inspects the data in it and records the decision, then routes the request to the right provider or server.
Go deeper
Learn more
- Edge vs cloud AI gatewayA data sovereignty framework for deciding where LLM, MCP and A2A traffic should be enforced.
- Ferentin vs open-source AI gatewaysHow Ferentin compares with agentgateway, LiteLLM, Portkey and Bifrost on identity, approvals and audit.
- What the LiteLLM supply chain attack revealedWhy AI gateway credentials need envelope encryption and edge-local decryption.
Related
- MCP SecurityDecide which agents reach which MCP servers and tools, and inspect every call.
- Enterprise-Managed AuthorizationYour IdP approves each agent connection. Ferentin authorizes every call that follows.
- AI Data Loss PreventionRedact or block secrets and personal data in prompts, tool calls and results.
- EndpointsA managed macOS agent that finds and governs the AI tools on every Mac.
See it on your own stack
Walk through it with our team on your identity provider, your models and your MCP servers. Or start free today.