MCP
Connect a workbench to a Model Context Protocol server
The MCP tool connects a workbench agent to a Model Context Protocol server that no other Plural integration covers. Plural connects to the server over the streamable HTTP transport, lists the tools it exposes, and offers each of them to the agent with the name, description, and input schema the server publishes.
The tool has two parts. An MCP server stores the endpoint URL and authentication and can be shared by several configured tools. The MCP tool references one server and decides which workbench subagents can use its tools. For the complete list of integrations and general tool behavior, see Workbench tools.
Prerequisites
Before you begin, make sure you have:
- An MCP server that supports the streamable HTTP transport. Servers that only offer the legacy HTTP+SSE transport are not supported.
- Network access from wherever Plural Console runs to the server's endpoint.
- The server's credentials, if it requires authentication: static header values such as an API key, or OAuth 2.0 client credentials for a token endpoint.
- Permission in Plural Console to create a configured tool and edit the target workbench.
- An existing workbench, or permission to create one. See Setting up a workbench.
Use credentials for a dedicated integration identity with only the access the agent needs. The agent can call every tool the server lists, with any arguments the server accepts.
Capabilities
The agent's operations are the tools the MCP server exposes. Plural adds no operations of its own. Each server tool appears to the agent as mcp_<tool name>_<server tool>, where <tool name> is the configured tool's name and <server tool> is the name the server reports.
Plural lists the server's tools when a job starts. Tools the server adds later become available to the next job. When the agent calls a tool, Plural forwards the arguments unchanged and returns the text content of the result to the agent. Non-text content, such as images or embedded resources, is dropped.
Plural does not limit which of the server's tools the agent can call. If a server exposes tools that change data, scope its credentials so those changes are acceptable, or deny calls to those tools by name with a workbench policy.