Appearance
MCP Server Connection
The Model Context Protocol Server Connection connects a Project to a remote MCP server. It can make the server's tools available to supported AI Agent functionality and can invoke MCP tools through supported Connection commands.
Only connect to an MCP server that you trust. Its tool descriptions, results, and side effects originate outside the Project.
Capabilities
| Capability | Availability |
|---|---|
| Connection service identifier | mcp |
| Page Flow commands | List MCP Tools, Call an MCP Tool |
| Cloud Flow commands | List MCP Tools, Call an MCP Tool |
| Dedicated Server API | Not provided |
| AI Agent MCP support | Supported |
Client-supported commands are executed by the server. Exposing a tool through Page Flow does not move the MCP connection or its authentication into the browser.
Preparing the MCP Server
Obtain:
- The remote server URL.
- The supported transport:
sseorhttp. - Any bearer token, custom authorization value, or required headers.
- The tool names and input schemas expected by the server.
- The server's timeout and network requirements.
The Connection supports remote HTTP-based transports. It does not configure a local stdio MCP process inside the Project.
Profile Settings
| Setting | Purpose |
|---|---|
| Authorization Type | Select none, Bearer, or Custom. |
| Secret | Supplies the token or custom authorization value. |
| Additional Request Headers | Adds authentication or protocol headers represented as JSON. |
The Profile form also displays a user-name field inherited from general authentication configuration, but Basic authorization is not an available MCP authorization mode. Configure authentication according to the selected mode and server requirements.
Connection Settings
| Setting | Purpose |
|---|---|
| URL | HTTP or HTTPS URL of the MCP server. |
| Transport | sse or http. |
| Additional Request Headers | Connection-specific headers represented as JSON. |
| Timeout | Milliseconds before a request times out; the default is 10,000 milliseconds. |
Commands
| Command | Purpose |
|---|---|
| List MCP Tools | Returns the tools advertised by the connected MCP server, with optional pagination. |
| Call an MCP Tool | Invokes a named tool with a data object matching that tool's input schema. |
Tool availability can change when the remote server changes. Do not hardcode assumptions about a tool without handling missing tools and schema changes.
Access
The Connection exposes a Call Method Access operation. Public access is permitted but produces a warning.
Avoid public access unless the Project validates the requested tool name and arguments. An unrestricted tool call can expose data or trigger external actions using the MCP server's authority.
AI Agent Use
When a supported AI Agent configuration uses an MCP Connection, review every advertised tool before making it available to the agent. Consider:
- Whether the tool reads or changes external data.
- Whether the tool accepts untrusted text or URLs.
- Whether the MCP server applies its own authorization.
- Whether the agent should have access to all tools or only a restricted server.
- What information a tool result can reveal to the user.
Treat tool output as untrusted external content. The MCP server can return errors, instructions, links, or data that should be validated before downstream use.
Testing
First use List MCP Tools to confirm connectivity and authentication. Then call a low-risk tool with controlled input before enabling the Connection in an AI Agent or user-facing workflow.
Test unavailable tools, invalid parameters, authorization failures, timeouts, and malformed server responses.