Model Context Protocol
A hosted MCP server for the tools your business already runs on.
MCP is the standard that lets an AI client call your software. FloConnector is the server side of it, run for you: authorise a platform once, choose which tools to expose, and point any MCP client at the endpoint. Nothing to deploy, nothing to keep running.
- 58connectors
- 6,533tools
- 24categories
Works with your client
One endpoint. Any MCP client.
A FloConnector profile is a standard MCP server, not an integration written for one assistant. The same URL works wherever you add it. These three have a walkthrough filmed end to end; a client we have not filmed still connects the same way.
ChatGPT MCP
ChatGPT reaches your business tools through the same profile endpoint Claude uses.
Connect ChatGPT
Mistral MCP
Le Chat supports MCP connectors, so it plugs into the same profile endpoint as every other client.
Connect Mistral Le ChatHosted, not local
Most MCP servers are something you run. This one is not.
The usual MCP setup is a package you install, a config file you edit and a process that has to be alive on the machine you are sitting at. That is fine for a developer wiring up a local database. It is the wrong shape for a business that wants its accounting, its CRM and its inbox reachable from whichever device someone happens to be on.
What an MCP server actually is-
Credentials stay on our side
Each vendor token is encrypted before storage and refreshed server-side. The AI client never sees one.
-
You choose the tools
A profile exposes only what you switch on, so a connection can be read-only, or scoped to one job.
-
Access is per person
Profiles are assigned to people. Someone leaving does not take a working endpoint with them.
-
Nothing to keep alive
No local process, no npm install, no machine that has to be awake for the connection to work.
The thing nobody warns you about
More tools is not better.
Connect enough MCP servers and the model gets worse, not better. Every tool definition costs context before a question is even asked, and a model handed hundreds of choices picks the wrong one more often.
The profile is our answer: a chosen set of tools for a chosen job rather than everything you own, and you can run several.
Read how we handle tool overloadQuestions, answered
What is a hosted MCP server?
An MCP server exposes one system's data and actions in a language an AI client understands. A hosted one runs on someone else's infrastructure, so you point your client at a URL instead of installing a package and keeping a process alive on your own machine.
Which AI clients does FloConnector work with?
Any client that can add a remote MCP server over HTTP. We publish walkthroughs for Claude, ChatGPT and Mistral Le Chat because those are the ones we have filmed end to end, but the endpoint is a standard MCP server and is not written against any one of them.
Do I have to build my own MCP server?
No, and for most business tools it is not worth it. You would be writing and hosting an OAuth integration, a token refresh path and a tool schema per vendor, then keeping all of it working as their APIs change. That is the work FloConnector does.
Does the AI see my credentials?
No. Each vendor credential is encrypted before it is stored and FloConnector makes the API calls itself. The client sees the tools and their results, never the credential behind them.
What stops the AI touching something it should not?
A profile carries only the tools you switch on. You can expose reads and leave every write tool off, and you can run separate profiles for separate jobs so one connection is not a key to everything.
Is one endpoint enough for several tools?
Yes. A profile can carry connections to many platforms at once, so a single endpoint in your client reaches whichever of them you have enabled.