MCP gateway

Agents use your tools only where you allow.

Each agent acts as the person who asked, reaches only the tools on its list, and every call is logged.

About 3 weeks
to live on real traffic
Founder-led
no handoff to juniors
Fixed price
agreed before work starts
Tool calls Every agent call to a company toolLive
Calls 38,204 today
TimeAgentToolActionResult
09:14:19Meeting notesCalendarread eventsAllowed
09:14:05Sales assistantCRMupdate deal stageHeld for approvalAllowed once
09:14:03Company searchDocssearch pagesAllowed
09:14:02Support triageTicketsread ticketAllowed
09:13:58Code reviewReposread pull requestAllowed
09:13:55Invoice matchingERPread purchase orderAllowed
09:13:51Support triageEmailsend outside AcmeBlocked
09:13:47Text to SQLWarehouserun read queryAllowed
09:13:44Company searchDrivelist filesAllowed
09:13:40Code reviewReposcomment on lineAllowed
09:13:36Support triageTicketsadd internal noteAllowed
Approvals none waiting

Writes that need a person show here before they run.

Approval needed Waiting Allowed
Agent
Sales assistant
Tool
CRM, update deal stage
Change
Deal 4471 Proposal to Won
Approver
Sales ops lead
DenyAllow once
Policy for every agent
Read any connected toolAllowed
Write to CRM or ERPNeeds approval
Send email outside AcmeBlocked
Delete recordsBlocked
The problem

Every team connects its agents to tools its own way

Nobody approved the list, the agents act with shared service keys, and when a record changes nobody can say which agent changed it or for whom.

80%

of companies say their AI agents took actions nobody intended.

Sourcesilicon.co.uk

44%

of organizations have a policy to secure their AI agents.

Sourcesilicon.co.uk

43%

of tested tool servers had command injection flaws.

Sourceequixly.com

How we fix it

No agent touches a tool until the allow-list says it may.

What changesWithout itWith it
ToolsEach team adds its own servers.One list, approved by an owner.
AccessAgents act with a shared key.Each call runs as the person asking.
AuditNo record of what an agent did.A log entry for every call.
How it works

What we build

What changes for your teams. We pick the tools to fit the systems you already run.

Calls 38,204 today
TimeAgentToolActionResult
09:14:19Meeting notesCalendarread eventsAllowed
09:14:05Sales assistantCRMupdate deal stageHeld for approvalAllowed once
09:14:03Company searchDocssearch pagesAllowed
09:14:02Support triageTicketsread ticketAllowed
09:13:58Code reviewReposread pull requestAllowed
09:13:55Invoice matchingERPread purchase orderAllowed
09:13:51Support triageEmailsend outside AcmeBlocked
09:13:47Text to SQLWarehouserun read queryAllowed
09:13:44Company searchDrivelist filesAllowed
09:13:40Code reviewReposcomment on lineAllowed
09:13:36Support triageTicketsadd internal noteAllowed
What changes

Measured against where you started

Example: Acme Inc

What changes at a company like Acme Inc

Illustrative figures, not a client result. Your own baseline is taken in the first week.

MeasureBeforeAfter
Tool servers with a named ownerUnknownAll of them
Agent tool calls loggedNoneEvery call
Calls made with a shared keyMostNone
Time to approve a new toolNo review at allOne review, about 2 days

How others built it

Another company built the same kind of system and wrote publicly about how it went.

What you keep

Everything runs in your cloud, and stays yours

You own the config, the tests and the runbooks, so you can swap any part later.

The gateway running in your cloud, configured as code in your repo.

Policy for every agent
Read any connected toolAllowed
Write to CRM or ERPNeeds approval
Send email outside AcmeBlocked
Delete recordsBlocked

An allow-list of approved tools, each with a named owner.

Calls 38,204 today
TimeAgentToolActionResult
09:14:19Meeting notesCalendarread eventsAllowed
09:14:05Sales assistantCRMupdate deal stageHeld for approvalAllowed once
09:14:03Company searchDocssearch pagesAllowed
09:14:02Support triageTicketsread ticketAllowed
09:13:58Code reviewReposread pull requestAllowed
09:13:55Invoice matchingERPread purchase orderAllowed
09:13:51Support triageEmailsend outside AcmeBlocked
09:13:47Text to SQLWarehouserun read queryAllowed
09:13:44Company searchDrivelist filesAllowed
09:13:40Code reviewReposcomment on lineAllowed
09:13:36Support triageTicketsadd internal noteAllowed
  • Sign-in, so every call runs as the person asking.

  • A log of every tool call, and the tests that prove the rules.

  • A runbook your platform team owns.

Typical building blocks: agentgateway with the official MCP Registry schema; OAuth 2.1 and the stateless 2026-07-28 spec; tools off until enabled.

Alternatives: IBM ContextForge; Cloudflare MCP portals.

How the weeks run

Live in about 3 weeks

The work happens in your cloud, with the people who will own it.

  1. Week 1Find every tool server your agents reach and who owns it.
  2. Week 2Put the gateway in your cloud, connect sign-in and move the first agents onto it.
  3. Week 3The allow-list, the logs and the approval step live, then hand over the runbook.
  4. After thatRunning on real traffic

    Agents reach tools only through an allow-list.

Before we start

  • Admin access to one cloud account and to your identity provider.
  • A list of the tool servers your agents use, or an hour with whoever knows.
  • A named owner for the allow-list, about 3 hours a week.
Fit

Is this the right first build?

It fits when

  • Agents connect to tools nobody approved.
  • Nobody can say what an agent did.
  • A team on your side will own it after we leave.

It does not fit when

  • You want a hosted tool picked and resold to you.
  • You want a strategy deck in place of working software.
  • Nobody on your side can own it after we leave.

What’s included and what isn’t

Included

  • One gateway in one cloud account.
  • Up to 10 existing tool servers behind the allow-list.
  • Sign-in through one identity provider.
  • Up to 3 agents moved onto it.

Not included

  • Writing new tool servers.
  • Fixing flaws inside the tool servers themselves.
  • A second identity provider.
Fixed quoteOne price, agreed before work starts

About 3 weeks. You get a price range before the call and one fixed number after it. That number is the invoice.

  • If the work takes longer than we planned, we absorb the cost.
  • You pay 50% at the start and 50% at acceptance. Builds of 6 weeks or more pay 40% at the start, 30% at the midpoint and 30% at acceptance.
  • If a build starts within 60 days of your Audit results, the full Audit fee comes off its price.
  • We start three new projects a quarter, so your start date depends on when you book.
Book a demoSee starting prices
FAQ

Questions teams ask before they book

Which gateway do you use?

We pick the one that fits the systems you already run. You own the config, the tests and the runbooks, so you can swap any part later. Most often agentgateway with the official MCP Registry schema; IBM ContextForge or Cloudflare MCP portals when they fit better.

Does it run in our cloud?

Yes. The gateway, its logs and its list of approved tools stay in your cloud account.

What does acting as the user mean?

The agent signs in as the person who asked, through OAuth 2.1, so it can reach only what that person may reach.

Do our existing tool servers still work?

Yes. They sit behind the gateway, and agents change only the address they call.

How is a new tool approved?

Someone asks for it, a named owner reviews it and it is switched on in the config. Every tool starts switched off.

Do we need the LLM gateway first?

No. They work well together: the LLM gateway governs model calls and this one governs tool calls.

MCP gateway, running in your cloud in about 3 weeks.

Agents reach tools only through an allow-list. You own the config, the tests and the runbooks.