Skip to main content
RoutMe

Connect your systems: API, database, MCP

RoutMe connects the agent to your systems in three ways, all read-only. Import an OpenAPI spec and pick the endpoints that become tools. Connect a PostgreSQL or MySQL database and approve validated SELECT queries. Or register an MCP server. There is no glue code to write, and each connection can be tested before it goes live.

RoutMe chat widget on a SaaS help page answering a plans question with a pricing table and a plan picker
A table drawn from connected plan data.

What you get with connect your systems

  • API connections

    Paste a spec or give its https URL. Choose none, an API key, a bearer token, a username and password, or OAuth 2 client credentials. Secrets are write-only after saving.

  • Database connections

    PostgreSQL and MySQL over TLS. Generate lookup or search tools from your tables, or write your own SELECT with named parameters. Every tool is validated and testable.

  • MCP servers

    Register any MCP server by URL, run a connection test that lists its tools, and let a widget use only the servers you select. Only tools marked read-only are used.

  • Answers about one person, safely

    Pin a parameter such as customerId to the verified visitor. RoutMe fills it on the server and hides it from the agent, so nobody can ask for someone else's records.

How it works

  1. 1

    Add a connection

    Import the spec, enter the database details with a read-only user, or point at your MCP server. Private and internal addresses are refused.

  2. 2

    Choose the tools

    Each read-only endpoint or approved query becomes a tool. Write a clear description of what it returns and when to use it. Endpoints that change data can be viewed but not turned on.

  3. 3

    Test with the real thing

    The Test button calls your actual API or database with the saved settings and shows the shaped result, with credentials masked.

  4. 4

    Choose where it is used

    In the widget editor, give a widget every active connection or only the ones you tick. A widget with nothing selected has no external tools at all.

Examples

The same capability across industries

Clinic
A database tool looks up a patient's next appointment by a pinned patient id, and returns only the columns on the allow list.
Travel
An imported booking API serves itinerary and baggage rules. The agent never sees an internal notes field because it is denied.
Real estate
A search tool filters listings by area, bedrooms and price from the listings database.
SaaS
An MCP server exposes the ticket system, so a customer can ask for the status of their own request.
Banking
A read-only API returns account details for the signed-in customer. Emails and phone numbers are masked.
Retail
An orders API and a stock database answer where an order is and whether a size is in stock.

Limits worth knowing

  • Everything is read-only. Anything other than GET, HEAD or OPTIONS cannot be turned on, and only SELECT statements run.
  • RoutMe connects over the public internet with TLS. Private networking is not available yet.
  • API calls time out after 10 seconds by default, and responses are cut at 256 KB.
  • Your own API should still check permissions, and the credential you give RoutMe should read only what is needed.

Documentation

Frequently asked questions

Do I need to write code to connect an API?

No. You import an OpenAPI or Swagger description and choose which endpoints become tools. If your API has no spec, an MCP server is the other route.

Can the agent write to my database?

No. Only single SELECT statements run, with values passed as parameters. You should also use a database user that can only read the tables you expose.

How do I keep sensitive fields out?

Deny fields so they are removed before the agent sees the response, or allow only certain fields. You can also mask emails and phone numbers and cap how much comes back.

What happens when my API changes?

Refresh the spec. RoutMe lists what was added, removed or changed, and new tools start switched off.

Can other AI apps use the same tools?

Yes. Each API and database connection also exposes a standard MCP endpoint with a token shown once. It serves only enabled read-only tools.

See it on your own data

Tell us what your customers ask. We will show the agent answering from your systems.