1. Try the sandbox
- Attach MCP Client Tool to your AI Agent’s tool connection.
- Set Endpoint to
https://plainrouter.com/mcp/sandbox. - Select Server Transport → HTTP Streamable and Authentication → None.
- Set Tools to Include → Selected and choose
get_account_stateandget_signal_health. - Run a manual chat with the sandbox prompt below. Name the node
plainrouter-testto distinguish it.
sandbox: true. This checks the connection, not production data or ad execution.
2. Create a workspace key
In Plainrouter, select your workspace and open Settings → Workspace keys. Create a Read key for analysis; use Write for Actions, including Actions record reads. Copy it once into your protected credential configuration. Each key covers one workspace. Never put keys in prompts or source control. Key permissions and replacement.3. Add Plainrouter to the client
Duplicate or reconfigure the MCP Client Tool for production:
Name the production node
plainrouter, attach it to the AI Agent, and keep credentials in n8n’s credential store. Add other tools only as needed.
4. Verify the connection
Start with this read-only request:Send conversions from a workflow
Conversion sending uses a separate HTTP request toPOST https://plainrouter.com/api/v1/events, authenticated with a Server secret, not an MCP Workspace key. Map the Conversion API body from a real eligible website event. Preserve its event ID and original time on retries, and supply the required consent and visitor context.
Start with the automation solution and synthetic n8n exercise. A sandbox result does not verify production event delivery or attribution. Verify HTTP event acceptance and Meta delivery separately from the MCP connection.
5. Troubleshooting
- Only an SSE Endpoint field appears: that is the older node shape. Use a current MCP Client Tool with HTTP Streamable support.
- 401 on calls: use the raw Workspace key in Bearer Auth; do not double-prefix it or use a Server secret.
- Agent never calls the tool: check the tool connection and selected-tool list, then inspect the workflow execution rather than trusting the chat summary.