This Agent Plugin connects GitHub Copilot to the remote MCP server included with Puppetflow. It lets an agent execute workflows that a user has explicitly exposed while credentials, authenticated sessions, execution state, and audit history remain on the Puppetflow server.
This is not a replacement for Playwright. Use the browser tooling built into the Copilot harness for ad hoc navigation and development testing. Use Puppetflow for approved, repeatable processes that need server-held credentials, durable runs, human approval, structured output, or auditable artifacts.
- A reachable Puppetflow instance
- MCP enabled for the target workspace
- At least one flow marked Available in MCP
- A Puppetflow MCP token
Copy the HTTP endpoint shown in Workspace > Settings > Instance MCP, then configure these environment variables before starting Copilot:
export PUPPETFLOW_MCP_URL="https://your-instance.example/api/mcp-server/http"
export PUPPETFLOW_MCP_TOKEN="mcp_your_token_here"The token is sent only to the configured Puppetflow endpoint as a Bearer authorization header. The plugin does not collect telemetry or send data to another service.
The exact tools are controlled by the Puppetflow workspace administrator. They can include:
- searching and inspecting exposed flows;
- running flows with structured JSON input;
- following durable runs and retrieving structured output;
- continuing explicit human validation checkpoints;
- listing screenshots, downloads, and recordings;
- authoring flows when those tools are intentionally enabled.
See the Puppetflow MCP guide for server configuration and access controls.