Tinybird MCP server

View original on GitHub

An MCP server to interact with a Tinybird Workspace from any MCP client.

Features

Query Tinybird Data Sources using the Tinybird Query API
Get the result of existing Tinybird API Endpoints with HTTP requests
Push Datafiles

It supports both SSE and STDIO modes.

Usage examples

Setup

Installation

Using MCP package managers

Smithery

To install Tinybird MCP for Claude Desktop automatically via Smithery :


npx @smithery/cli install @tinybirdco/mcp-tinybird --client claude

mcp-get

You can install the Tinybird MCP server using mcp-get :


npx @michaellatman/mcp-get@latest install mcp-tinybird

Prerequisites

MCP is still very new and evolving, we recommend following the MCP documentation to get the MCP basics up and running.

You’ll need:

Configuration

1. Configure Claude Desktop

Create the following file depending on your OS:

On MacOS: ~/Library/Application Support/Claude/claude_desktop_config.json

On Windows: %APPDATA%/Claude/claude_desktop_config.json

Paste this template in the file and replace <TINYBIRD_API_URL> and <TINYBIRD_ADMIN_TOKEN> with your Tinybird API URL and Admin Token:


{
    "mcpServers": {
        "mcp-tinybird": {
            "command": "uvx",
            "args": [
                "mcp-tinybird",
                "stdio"
            ],
            "env": {
                "TB_API_URL": "<TINYBIRD_API_URL>",
                "TB_ADMIN_TOKEN": "<TINYBIRD_ADMIN_TOKEN>"
            }
        }
    }
}

2. Restart Claude Desktop

SSE mode

Alternatively, you can run the MCP server in SSE mode by running the following command:


uvx mcp-tinybird sse

This mode is useful to integrate with an MCP client that supports SSE (like a web app).

Prompts

The server provides a single prompt:

tinybird-default : Assumes you have loaded some data in Tinybird and want help exploring it.
- Requires a “topic” argument which defines the topic of the data you want to explore, for example, “Bluesky data” or “retail sales”.

You can configure additional prompt workflows:

Create a prompts Data Source in your workspace with this schema and append your prompts. The MCP loads prompts on initialization so you can configure it to your needs:


SCHEMA >
    `name` String `json:$.name`,
    `description` String `json:$.description`,
    `timestamp` DateTime `json:$.timestamp`,
    `arguments` Array(String) `json:$.arguments[:]`,
    `prompt` String `json:$.prompt`

Tools

The server implements several tools to interact with the Tinybird Workspace:

list-data-sources: Lists all Data Sources in the Tinybird Workspace
list-pipes: Lists all Pipe Endpoints in the Tinybird Workspace
get-data-source: Gets the information of a Data Source given its name, including the schema.
get-pipe: Gets the information of a Pipe Endpoint given its name, including its nodes and SQL transformation to understand what insights it provides.
request-pipe-data: Requests data from a Pipe Endpoints via an HTTP request. Pipe endpoints can have parameters to filter the analytical data.
run-select-query: Allows to run a select query over a Data Source to extract insights.
append-insight: Adds a new business insight to the memo resource
llms-tinybird-docs: Contains the whole Tinybird product documentation, so you can use it to get context about what Tinybird is, what it does, API reference and more.
save-event: This allows to send an event to a Tinybird Data Source. Use it to save a user generated prompt to the prompts Data Source. The MCP server feeds from the prompts Data Source on initialization so the user can instruct the LLM the workflow to follow.
analyze-pipe: Uses the Tinybird analyze API to run a ClickHouse explain on the Pipe Endpoint query and check if indexes, sorting key, and partition key are being used and propose optimizations suggestions
push-datafile: Creates a remote Data Source or Pipe in the Tinybird Workspace from a local datafile. Use the Filesystem MCP to save files generated by this MCP server.

Development

Config

If you are working locally add two environment variables to a .env file in the root of the repository:


TB_API_URL=
TB_ADMIN_TOKEN=

For local development, update your Claude Desktop configuration:


{
  "mcpServers": {
    "mcp-tinybird_local": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/your/mcp-tinybird",
        "run",
        "mcp-tinybird",
        "stdio"
      ]
    }
  }
}

Published Servers Configuration


"mcpServers": {
  "mcp-tinybird": {
    "command": "uvx",
    "args": [
      "mcp-tinybird"
    ]
  }
}

Building and Publishing

To prepare the package for distribution:

Sync dependencies and update lockfile:


uv sync

Build package distributions:


uv build

This will create source and wheel distributions in the dist/ directory.

Publish to PyPI:


uv publish

Note: You’ll need to set PyPI credentials via environment variables or command flags:

Token: --token or UV_PUBLISH_TOKEN
Or username/password: --username/UV_PUBLISH_USERNAME and --password/UV_PUBLISH_PASSWORD

Debugging

Since MCP servers run over stdio, debugging can be challenging. For the best debugging experience, we strongly recommend using the MCP Inspector .

You can launch the MCP Inspector via npm with this command:


npx @modelcontextprotocol/inspector uv --directory /Users/alrocar/gr/mcp-tinybird run mcp-tinybird

Upon launching, the Inspector will display a URL that you can access in your browser to begin debugging.

Monitoring

To monitor the MCP server, you can use any compatible Prometheus client such as Grafana . Learn how to monitor your MCP server here.