Sinfonia MCP: talk to your RPA operation

Instead of opening screens, filtering lists and hunting for logs, how about just asking “what failed today?”? With the launch of Sinfonia MCP, you connect your Sinfonia Account to your favorite AI assistant (Claude, Codex, Gemini, Cursor, VS Code and others) and query or operate your automation by chatting.

The server is already available at https://mcp.sinfonia.live/mcp, at no extra cost and with no token to configure.

The day-to-day problem

Anyone who runs an RPA operation knows the routine. A bot failed overnight and, to understand what happened, you have to open the dashboard, filter the failed executions, open them one by one, read the end of the log, check whether the agent was online and, finally, decide whether a re-run is worth it.

None of these steps is hard. The problem is adding them up: lots of clicks to answer simple questions, repeated every day, by several people. And questions like “why hasn’t the payroll bot started yet?” require cross-checking information from different screens (queue, triggers and agents).

Concepts, without the jargon

AI assistants don’t know your data

An assistant like Claude or ChatGPT is great at understanding natural-language questions and explaining things. But on its own it knows nothing about your bots: it has no access to your executions, your agents or your logs. To answer questions about your operation, it needs a secure way to fetch that data.

What is the Model Context Protocol

The Model Context Protocol (MCP) is an open standard that defines how an AI assistant connects to external systems. Think of it as a USB-C port for AI: a system implements the protocol once and works with any compatible client.

An MCP server offers the assistant two kinds of resources:

• Tools: well-defined actions the assistant can call, such as “list yesterday’s failed executions” or “run bot X”.
• Prompts: ready-made scripts for common tasks, such as “diagnose today’s failures”.

The assistant decides which tools to call to answer your question, combines the results and explains them in natural language. When a tool changes something, the client asks for your confirmation first.

Where Sinfonia MCP fits

    flowchart LR
    U[You] -->|question| A[AI assistant]
    A -->|calls tools| M[Sinfonia MCP]
    M -->|with your user and Profile| S[Sinfonia]
    S -->|data| M
    M -->|results| A
    A -->|answer| U
  

Sinfonia MCP sits between the assistant and the platform. Everything it does is with your user and within the permissions of your Profile.

What you can do

• Query bots, executions, queue, agents and triggers.
• Diagnose failures: the assistant reads the error and the end of an execution’s log and explains what happened.
• Monitor the operation: success rate over the last days and the state of the agents.
• Operate: run a bot and pause, resume or stop an execution, always at your request.

The full list of tools, parameters and permissions is in Tools and permissions.

How it makes your day easier

Morning triage. Whoever starts the support shift asks for a summary of the operation: online and offline agents, queue, running executions and recent failures. The ready-made resumir_operacao prompt does this in a single message.

Failure diagnosis. Instead of opening log after log, ask “why did execution 4512 fail?”. The assistant reads the error and the end of the log and points to the likely cause, such as a timeout on the portal or a field whose identifier changed. With the diagnosticar_falhas prompt, it groups all failures in the period by cause.

Before publishing a version. The RPA developer uses the investigar_bot prompt to see a bot’s versions, triggers and recent history and understand how it has been behaving.

Visibility for management. “What was last week’s success rate?” becomes a direct answer, with no report to build.

On call. A transient failure at night? Ask to re-run the bot on the right agent without opening the platform, with confirmation before the action.

Usage examples

You: What failed yesterday?

Assistant: Yesterday there were 3 failed executions, all of bot-notas-fiscais on agent AG-01. The cause was the same in all three: TimeoutError when accessing the city hall portal, after 120 seconds. The following executions, from 2 PM on, finished successfully.

You: Run it again on AG-01.

Assistant: I queued execution 4530 of bot-notas-fiscais (version 1.4.0) on AG-01.

You: Why hasn’t the payroll bot started yet?

Assistant: bot-folha is 3rd in the AG-02 queue, but that agent has been offline since 7:40 AM. The two executions ahead of it are also waiting for the same agent. Check the AG-02 machine or run the bot on another available agent.

Other questions that work well:

• “Which agents are offline?”
• “List the schedules of the reconciliation bot.”
• “Which bots failed the most in the last 7 days?”
• “Stop execution 4512.”

See more in Usage examples.

Security and control

Connecting an AI to your operation requires care, and Sinfonia MCP was designed with that in mind:

• Follows your Profile: MCP has no permissions of its own. The assistant only sees and does what you can already see and do on the platform.
• Browser login: authentication uses OAuth 2.1. You type your password and verification code on Sinfonia’s pages, never in the AI client, and there is no token or key to store.
• Actions only at your request: write tools are used only when you explicitly ask, and most clients ask for confirmation first. Read the request before approving.
• Logs are data: instructions written inside an execution log are not followed by the assistant.
• Time-limited session: the session expires in 24 hours, and you can revoke access at any time by removing the server from the client.

What it doesn’t do

MCP is focused on day-to-day querying and operation. Creating, editing or publishing bots and versions, creating or changing triggers and managing users, tokens, plugins and environment variables are still done on the platform.

Getting started

1. Add the https://mcp.sinfonia.live/mcp server to your AI client.
2. Log in with your Sinfonia user and enter the code sent by email.
3. Authorize access and ask your first question.

The step-by-step for each client is in the documentation:

Next steps

This is the first step in bringing AI into RPA operations. We want to hear how you are using Sinfonia MCP and what you would like it to do next. Not using Sinfonia yet? Request access and start talking to your operation.