MCPswitch

For companies with an API

Let AI assistants use your product.

MCPswitch turns your API into an MCP server, the standard connector AI assistants use to take real actions. We design it, test it and host it for you.

Start freeHow it works

Assistant
  1. Book me a 30-minute call with Dana next Tuesday afternoon.
  2. Acme Booking find_open_slots
  3. Dana is free Tuesday at 2:00 pm or 3:30 pm. Which one works?
  4. 2:00, please.
  5. Acme Booking create_booking
  6. Done. You're booked with Dana on Tuesday at 2:00 pm, and she has the invite.

Example. Acme Booking is a made-up company.

MCP is how AI assistants use other products.

AI assistants no longer just answer questions. They book, look up, update and send, using tools other companies provide. The Model Context Protocol (MCP) is the open standard for those tools.

An MCP server is your product's connector. With one, a customer can ask their assistant to do something in your product, in plain words, and it happens.

  1. Someone asks"Book me a call with Dana on Tuesday."
  2. The assistant picks a toolIt sees your tools and chooses the right one.
  3. Your API does the workThe booking is made in your system, like any other.

Why companies use MCPswitch

You already have an API. We turn it into something assistants can use well, and we keep it running.

Your product, inside the assistant

Customers ask in their own words and get things done in your product, without opening it. It's a new front door to what you already built.

Tools that fit real tasks

Turning every endpoint into a tool gives an assistant hundreds of options and it guesses wrong. We pick the handful of actions your customers actually need, and name and describe them so assistants choose correctly.

  • No MCP expertise needed. You bring the API. We handle the protocol, the schemas and the hosting.
  • You decide what's allowed. Read only, or writes too. Risky actions wait for a person to approve them.
  • Your keys stay locked. API credentials are encrypted and never shown to the AI or written to logs.
  • Hosted and kept working. We run the server, watch your API for changes, and version every update so you can roll back.

How it works

  1. 1
    Share your API

    Paste a link to your OpenAPI or Swagger file, or to your API docs. We read every endpoint and tell you what we can support.

  2. 2
    Say what assistants should do

    In a sentence or two. "Customers can check availability, book and cancel." We ask a few questions where it matters.

  3. 3
    Review the tools

    We design a small set of tools and explain each one. Switch any off, rename them, and choose which need your approval.

  4. 4
    Test, publish, share

    Try it with sample data first, then live. Publish and you get a server address to give your customers.

You stay in control

An assistant acting in your product should follow your rules. These are enforced on our server, not left to the AI to remember.

  • Approvals for risky actions. Refunds, cancellations and deletions wait for your OK. The approval covers that exact request, once.
  • Keys for every caller. Give each customer or partner their own key, limited to certain tools, a daily allowance and an expiry.
  • Customers see only their own records. A key can be tied to one customer, and anything that isn't theirs stays hidden.
  • A full activity log. Every call, its result and how long it took, with no personal data stored in the log.
  • An off switch. Pause the server or revoke a key and the next request is refused.

Everything it takes to run it

Not just a generated file. A service that's designed, tested and looked after.

  • Hosted serverYour own address, running on our infrastructure.
  • Preview serverTry every tool with sample data before anything touches your real API.
  • Testing toolsRun each tool, check it against your docs, and see if an assistant picks the right one.
  • Versions and rollbackEvery change is a version. Go back in one click.
  • API change alertsWe check your API description daily and tell you what changed.
  • Encrypted credentialsAPI keys, tokens and OAuth client secrets, sealed at rest.

Free while we're in early access

We're working closely with the first companies to connect their APIs. Everything is included, with up to five projects per account. Paid plans come later, and we'll tell you well before anything changes.

Early access

$0

  • Up to 5 MCP servers
  • AI-assisted tool design
  • Hosting, keys, approvals and logs
  • Preview and testing tools
Start free

Questions

Do I need to change my API?

No. MCPswitch sits in front of the API you already have and calls it the way your own apps do.

What do you need from us?

An OpenAPI or Swagger description of your API (a link to your docs page often works too), a sentence about what assistants should do, and an API key or other credentials for the connection.

Which assistants can connect?

Any MCP client that supports remote servers over HTTP with a key, which includes the major AI coding tools and a growing number of chat assistants. We test against the official MCP software development kit. Tell us which assistant your customers use and we'll check it with you.

Is our API key safe?

It's encrypted on our server and only added to a request on its way to your API. The AI never sees it, and it's never written to logs. You can replace or revoke it at any time.

What if our API changes?

We check your API description every day. If something the server depends on changes, you're told what changed and which tools are affected. Nothing is changed without you.

What if our API isn't a good fit?

We'll say so, with the reason. GraphQL, SOAP, file uploads and acting as each individual user aren't in the automated setup yet. You can ask for a developer assessment of anything we can't automate.

What does it cost?

Nothing during early access. When paid plans arrive you'll hear about it first, and nothing is charged without your agreement.

Give AI assistants a way into your product.

You'll create an account next. Nothing is published until you say so.