Pluto Agent: No-Code Agents in Bubble Over API, and Why It's Not Agent 2 or MCP Beta

In early October 2026, three separate "agent" stories landed in the Bubble ecosystem within days of each other. I mixed them up at first. If you're shipping an MVP on Bubble with no-code and vibe coding, the AI news moves fast enough to blur together. And if you're an early-stage founder validating customer demand or getting ready for angel pitches, confusing these pieces costs you time and budget. Same topic: IronBase in Two Days on.

The Three "Agent" Stories in Bubble

1. Bubble Agent 2 (announced Oct 6, 2026). The generally available upgrade to Bubble's native in-editor AI assistant. It helps you design and configure your app inside Bubble, and it's billed in Bubble AI credits.
2. Bubble MCP beta (mcp.bubble.io/mcp). An official beta server that lets external AI coding environments, such as Claude Code or Codex, inspect and edit Bubble apps. Bubble's official documentation says working through Bubble MCP does not consume Bubble AI credits.
3. Pluto Agent (pluto-agent.com). A third-party no-code agent builder, introduced on October 12, 2026, by team member @keith21 in the Showcase category of the Bubble forum. It is not an official Bubble feature and not an in-editor builder. It's an external AI service you call from your Bubble app through the API Connector. Same topic: BuildPrint vs Codex CLI MCP.

(One more trap: docs.joinpluto.com is an unrelated marketing API product. It has nothing to do with pluto-agent.com.)

Integrating Pluto Agent Into a Solo MVP Stack

For a non-technical builder testing an MVP, Pluto follows the standard outbound API path:

1. Build the agent. Sign up on pluto-agent.com, configure your agent in Agent Studio, and generate an API key.
2. Configure the Bubble API Connector. Create an API collection with a private Authorization: Bearer [API_KEY] header.
3. Set up the endpoint. Define an action that calls POST https://api.pluto-agent.com/v1/response-agent.
4. Fill in the call parameters:
   • agent_id: the ID of your configured agent;
   • session_id: a unique session identifier, so the agent keeps track of each user's conversation;
   • client_prompt: the user prompt (make sure the JSON is valid; this is the easiest place to trip);
   • environment: test or live;
   • client_platform: set to bubble;
   • reset_session: a Boolean flag that clears conversation history when you need it.
5. Initialize. Initializing the call in the API Connector tests the connection and uses Pluto trial credits. After that, you can trigger the action from regular Bubble workflows.

Billing and Security Realities

• Two separate billing systems. Bubble AI credits only cover native Bubble features. Pluto Agent runs on Pluto credits, pay-per-call, priced on context length, reasoning, web search, and runtime duration. Run out and you get a 402 INSUFFICIENT_CREDITS response.
• Two different MCPs. Pluto has its own "MCP Connector" for attaching arbitrary MCP servers to your Pluto agent. That lives inside Pluto's tool layer. Bubble's official mcp.bubble.io/mcp server, the one that modifies Bubble applications, is a different thing.
• Be careful with data access. Pluto advertises tools like read-only Bubble database access and workflow triggering. For early MVP validation, I wouldn't grant broad Data or Workflow API access until you understand the privacy rules and the attack surface you're opening. Same topic: Agent 2 GA, MCP Beta,.

The Practical MVP Takeaway

I keep the stack modular. Test Pluto Agent only on the API Connector side with environment: test, and watch your runtime credits. Put your energy into getting real user feedback, and leave deeper production wiring until you have it.