MCP endpoint

Each workspace has one MCP endpoint. Every client and every agent connects to the same address. What each one sees depends on who it acts for.

https://mcp.flowplane.eu/<label>

The transport is MCP Streamable HTTP. Any client that supports remote MCP servers works.

Sign-in

In the browser (recommended). Clients that support MCP sign-in need only the address. On first use, your browser opens Flowplane: you sign in with your company account, choose what the client may do, and allow it.

With a key. For CI, scripts and servers, create a key on the connection in the console and send it on every request:

Authorization: Bearer <key>

A key belongs to one connection of one person. Treat it like a password.

Tools

The client sees the tools the agent may use, named like <app>_<operation>. When permissions or tools change, the list updates in clients that support it.

Your team’s notes on a tool are part of its description, so every agent follows the same house rules.

When a call is not allowed, the agent gets a short sentence saying why. When it needs approval, the agent gets a link, see Agents and access.

Compact mode

A workspace with ten apps easily has over a hundred tools, and many clients struggle with that. Add ?tools=compact to the address to get four tools instead: the agent searches for the tool it needs, reads its description, then calls it.

https://mcp.flowplane.eu/<label>?tools=compact

Every call is checked, approved and logged exactly like a direct call.

Code mode

?tools=code is compact mode plus one more tool that runs a short script of read calls. Use it for tasks that chain or count many reads, such as going through twenty mails or comparing two calendars. The agent gets back only the result, not every answer along the way.

https://mcp.flowplane.eu/<label>?tools=code

Scripts can only read, and each call in them is checked and logged on its own.