MCP (Model Context Protocol — an open protocol that lets a program expose tools to an agent) connects a service that is not in the Pailot catalog: an internal database, a custom script, a file directory.
You add an MCP server in workspace settings: name, address, and a key if the server requires one. After you save, its tools appear on the Agent block and on the MCP Tool block on the canvas.
How to connect a server
- Open workspace settings.
- Go to Deployed MCPs and add a server.
- Enter the name, server address, and an access key if needed. These are workspace form fields, not contour install variables.
- Save. Refresh fetches the latest parameter schema for each tool.
You can also add a server from the tool bar on an Agent block.
The built-in Browser block is a separate canvas step. It is not an MCP server that you add in workspace settings.
MCP Tool block
The block calls one selected tool on a server that is already connected. Argument fields come from that tool's schema.
- Agent — the model decides which tool to call.
- MCP Tool — a fixed step: you choose the server, the tool, and the arguments.
Incoming parameters
| Parameter | On the canvas | Type | Required | Description |
|---|---|---|---|---|
server | MCP Server | string | Yes | Server from workspace settings |
tool | Tool | string | Yes | Tool name on that server |
arguments | fields from the tool schema | object | Depends on the tool | Arguments the selected tool requires |
Outgoing parameters
| Parameter | Type | Description |
|---|---|---|
content | array | Tool response in the standard MCP format (an array of content parts) |
The next block reads the output as <mcp1.content>.
Permissions
| Action | Required permission |
|---|---|
| Configure MCP servers in the workspace | Admin |
| Call MCP tools from an agent | Write or Admin |
| See the tool list | Read, Write, or Admin |
| Run the MCP Tool block | Write or Admin |
Security
Do not grant a server more access than the step needs. The server address and key are workspace secrets. Do not publish them.
To expose your own workflow as an MCP tool, see Deploy Workflows as MCP.