docs

Flow MCP Servers

Integrate Plural AI with MCP to securely automate operations

Overview

Plural Flows natively integrate with Model Context Protocol (MCP) to become a portal to manage common operational tasks with AI. When registered, Flow chats can automatically call the MCP server via tool calling. Common use cases are:

  • Implementing ops tasks in a flow chat. Some examples: manually adding users to a custom payment plan, viewing basic account details, etc.
  • Exploring a read-only database connection with the Postgres MCP server
  • Querying data from JIRA, ServiceNow, etc. to enhance AI context
  • Converting Flows into your developer admin panel, with a cleaner security model than a no-code tool or a poorly maintained internal UI.

Anthropic has a large library of pre-built servers, and there's a rapidly growing open source community building more. In theory, the protocol could support any external data fetching needed to enhance an AI experience.

On top of the base value of MCP, Plural adds a few core governance features:

  • Audit logging of all MCP server calls
  • Confirmation workflows in case you want to require user confirmation before the AI initiates an MCP tool call
  • Authentication via JWT to end MCP servers, making it easy for developers to implement auth + authz and secure their server code
  • Permissions controlling who can access an MCP server within the Plural experience

Registering an MCP server

The YAML for this is relatively simple:

yaml
apiVersion: deployments.plural.sh/v1alpha1
kind: MCPServer
metadata:
  name: plural
spec:
  name: plural
  confirm: false
  url: https://plural-mcp.plural.sh
  authentication:
    plural: true
  bindings:
    write:
    - groupName: sre # only this group can bind this server to their Flows

From there, a developer with sufficient permissions can associate their Flow with the MCP server in the Plural UI by going to MCP Connections and clicking Change MCP Connections, like so: