BridgeratorWhat is an MCP server, and how do ChatGPT connectors work?
"Add an MCP server as a connector" sounds like jargon, but the idea is simple. A language model can only produce text. To do something — read a file, look up an order, run a command — it needs tools, and someone has to provide them. MCP is the standard way to provide tools to AI assistants.
Tools: how a model acts
A tool is a named action with a description and parameters, for example read_file(path) or start_process(command). When ChatGPT decides a tool would help, it doesn't run anything itself: it produces a structured request — "call read_file with path = D:\notes.txt" — and waits. Whoever owns the tool runs it and sends back the result. ChatGPT reads the result and continues: answers you, or calls another tool.
MCP: one plug for every tool
The Model Context Protocol is an open standard that fixes how this conversation happens. An MCP server is any program that speaks it. It tells the assistant "here are my tools and what they do", accepts tool calls and returns results. Because the protocol is shared, the same server works with any assistant that supports MCP — much like any USB device works with any computer.
There are MCP servers for databases, ticket systems, calendars, design tools — and for your own computer.
Connectors in ChatGPT
In ChatGPT, an MCP server you add yourself is called a connector. You create one in Settings → Apps & Connectors with Developer mode turned on: give it a name and the server's URL, and choose how to sign in. From then on, in any chat where the connector is enabled, ChatGPT can see that server's tools and use them. Developer mode availability depends on your ChatGPT plan.
Signing in: OAuth
A connector that touches private data must know whose data it is. That is what OAuth is for. When you add the connector, ChatGPT sends you to the server's sign-in page; you sign in there, and ChatGPT receives a token that lets it act for you — without ever seeing your password. Bridgerator uses OAuth 2.1 with PKCE, and changing your Bridgerator password revokes every token issued to ChatGPT at once.
The catch: ChatGPT lives in the cloud
A connector's URL must be reachable from the internet, because ChatGPT's servers call it. That is easy for a web service, but your PC sits behind a home router with no public address — and opening a port to it would be a bad idea.
Bridgerator solves this by splitting the MCP server in two:
- A public part at
https://that ChatGPT talks to. It handles sign-in and forwards tool calls; it doesn't store their contents.mcp.bridgerator.recondeck.com/ mcp - A bridge on your PC that opens an outbound secure WebSocket to the public part and actually runs the tools. Nothing connects in to your computer.
The tools themselves come from an open-source engine bundled with the bridge: files, search, terminal and processes. The bridge adds your rules on top — allowed and excluded folders, blocked commands, tools to hide.
What travels where
- You type a request in ChatGPT.
- ChatGPT calls a tool on the Bridgerator connector, with your OAuth token.
- The server checks the token, picks your PC and forwards the call over the bridge's connection (TLS).
- The bridge checks your rules, runs the tool and sends the result back the same way.
- ChatGPT receives the result and continues. The server keeps only a counter of how many calls you made — not what they were.
Note what this is not: it is not end-to-end encrypted. Traffic is encrypted in transit, and the server decrypts each call to route it — then forgets it.
In short
MCP is the plug, a connector is ChatGPT's socket, and Bridgerator is the cable to your Windows PC. Ready to try? Follow connecting ChatGPT to your PC — it takes about five minutes, and the Free plan includes 10,000 tool calls a month.