MCP gateway · deny by default

Give AI agents access to your systems — without opening a single port.

Talaiz is the secure, audited gateway between AI agents and your internal databases, APIs and tools. One outbound tunnel, per-tool authorization, human approvals, and a tamper-evident record of every call.

Start free See how it works

Apache-2.0 · self-host in minutes · no inbound firewall changes

agent → talaiz → your network

$ curl -s https://gateway/mcp \
  -H "Authorization: Bearer tlz_agent_…" \
  -d '{"method":"tools/call",
       "params":{"name":"billing__list_invoices"}}'

✓ allowed  policy: billing__list_invoices
✓ audited  sha256:9f3a…c21e
✓ redacted 4 sensitive fields

Outbound-only Human approval Tamper-evident audit

The problem

Connecting an agent to production is still an all-or-nothing bet.

Today teams pick one of two bad options — and both end the same way: an agent with more access than anyone can prove it should have.

Expose an internal service

Open a port, add a tunnel or a public endpoint. Now a production system is reachable from the internet, and every credential that reaches it is a breach waiting to happen.

Hand over a long-lived credential

Give the agent a database password or an API key. It works — until it leaks, or an agent does something no one authorized, and there is no record to explain what happened.

Talaiz inverts the connection: nothing listens, nothing is exposed, and every call is authorized and recorded.

How it works

One outbound tunnel. The gateway is the control point.

1
Install the connector
One small binary inside your network. It dials out to Talaiz and supervises your local MCP servers. No inbound port, no firewall change.
2
Declare the tools
The connector advertises the MCP servers and tools it can reach, so the gateway knows exactly what exists — nothing more.
3
Set policy per agent
Deny by default. Allow specific tools, require human approval, cap the rate — per agent, per tool, with a reason on every rule.
4
Audit every call
Who, what, when, with which arguments and which result — sealed into a tamper-evident hash chain you can verify on demand.

Why teams choose Talaiz

Security your auditors can actually read.

Deny by default

No tool is reachable unless a policy says so. Both the gateway and the connector enforce it, so one mistake cannot open a path.

Every call on the record

Each executed call produces exactly one audit event, sealed into a SHA-256 chain that verification re-checks end to end.

Outbound-only

The connector dials out. Nothing listens inside your network, so there is no new attack surface and no firewall ticket.

Human in the loop

Mark any tool require_approval and the call blocks until a person approves or denies it in the console.

Rate limits & redaction

Cap calls per minute before they leave the gateway, and strip sensitive fields before results reach an agent or the log.

Kill switch

Disable a connector or an agent and it takes effect within one policy push — seconds, not a redeploy.

Compliance

Evidence, not assurances.

Agent access is now in scope for the EU AI Act, GDPR, SOC 2, HIPAA and PCI DSS. Talaiz turns “we think it’s fine” into a per-call record you can hand to an auditor — and a control that stops the wrong call before it happens.

Read the compliance guide →

Deny-by-default policy enforcement
Per-call, tamper-evident audit trail
Human approval for sensitive tools
Redaction before results are stored
Kill switch within seconds

Put your agents on the record.

Start free, self-host, or talk to us about the managed cloud.

Start free See pricing