Skip to content

Turbo Notify in Langflow

Langflow reaches Turbo Notify through its MCP client: you add the Turbo Notify MCP server once in Langflow’s settings, and the MCP Tools component hands its tools to the agent in your flow. The agent can then send a message, reply, check whether a message was delivered and list your groups, always from your own number. There is no Turbo Notify component in Langflow’s Bundles list; the connection is the MCP server.

  • A Turbo Notify account with a connected number (quickstart).
  • A Langflow install with the MCP client, under Settings > MCP Servers. Connect steps 1 to 5 were run on Langflow 1.13.2.
  • An API key, created in API Keys (opens in a new tab) with the AI agent (MCP) template. Langflow’s server form has no browser sign-in, so this page uses the key.
  • A language model configured in Langflow for the agent.
  1. In Langflow, open Settings > MCP Servers and click Add MCP Server.

  2. Choose the Streamable HTTP/SSE tab.

  3. In Name, type turbo-notify. In Streamable HTTP/SSE URL, type:

    https://mcp.turbonotify.com/
  4. Under Headers, add the key Authorization with the value Bearer YOUR_KEY (the word Bearer, a space, then the key). Langflow masks the value once it is saved.

  5. Click Add Server. Langflow detects the transport by itself and lists the server with its tools (36).

  6. In a flow, add the MCP Tools component and pick the turbo-notify server. Leave Tool empty to give the agent every tool, or pick one tool to give it just that one. Turn on Tool mode and connect the component’s Toolset output to the Tools input of an Agent.

To keep the key out of the flow, the Langflow documentation describes a Credential global variable whose value includes Bearer and a space; you then type the variable’s name as the header value. Treat the key like a password.

Langflow signs in with the API key in the Authorization header. Scopes and keys are in MCP server.

A list of tools in Langflow does not prove that the key is valid. Turbo Notify shows its tool list to any caller. A key in the right format that is wrong (revoked, expired or deleted) only fails at the first tool call, with the error unauthorized: not_authenticated (“Turbo Notify did not accept the API key”). A value that is not in the format of a key (for example a mistyped key or the placeholder YOUR_KEY) is refused at once, with 401 invalid_token. If the agent’s first tool call fails like that, create the key again and replace the header. A value with another word before the key (for example Token or Basic) is treated as no key: the call fails with 401 unauthorized (“Sign in to use this server.”).

Run the flow in the Playground and ask: “List my WhatsApp numbers on Turbo Notify.” The agent calls list_numbers, which only reads, and answers with your number and whether it is connected.

To send, ask for a message to your own phone: “Send ‘test’ to +5511988887777 on WhatsApp.” The send is accepted at once and delivered a few seconds later; ask “Did it arrive?” and the agent reads the status with get_message_status.

The MCP server needs no update: the address is always the same. New tools show up the next time Langflow reads the server’s tool list.

  • The tool list is large: 36 tools, about 11 thousand tokens once they are sent to a model as function definitions. A model or plan with a small per-minute token limit can refuse the request: a free Groq plan with an 8,000 tokens per minute limit refused it. Pick one tool in the Tool field of the component, or use a model and plan with a larger limit.
  • The number sends one message at a time, spaced a few seconds apart. A flow that sends to a long list should send in smaller batches. See MCP server limits.
  • Sending to a group needs the Business plan or higher.
  • What was run: on 12 and 13 November 2026, on Langflow 1.13.2, the server was added with the key in the header, the 36 tools were listed, and list_numbers and send_text were called through Langflow’s own MCP client; the message was delivered. A model with tool calling (openai/gpt-oss-20b on a free Groq plan) was also given one of those tools, list_numbers, through the tools Langflow’s MCP client provides: it called the tool and answered correctly that the main number is connected. With all 36 tools the same request was refused for size. What was not run: the Playground and the Agent component, the MCP Tools component on the canvas, the key as a global variable, and Langflow Desktop or a hosted Langflow.