dimed-mcp 0.1.1

MCP server with Kedro prompts and tools (local stdio, zero-install via uvx).

Kedro MCP Server

An MCP (Model Context Protocol) server that helps AI assistants work consistently with Kedro projects.
It ships a tiny prompt and two read-only tools that return concise, versioned guidance for general Kedro usage and for converting a Jupyter notebook into a production-ready Kedro project.

---

🚀 Quick install (no setup, runs locally)

Prerequisites

• Python 3.10+
• uv (recommended)

VS Code – GitHub Copilot Chat (user or workspace settings):

{
  "github.copilot.chat.mcpServers": {
    "kedro": {
      "type": "stdio",
      "command": "uvx",
      "args": ["kedro-mcp@latest"],   // or pin: kedro-mcp==0.1.0
      "env": { "FASTMCP_LOG_LEVEL": "ERROR" },
      "disabled": false,
      "autoApprove": []
    }
  }
}

Claude Desktop / Cursor (generic MCP config):

{
  "mcpServers": {
    "kedro": {
      "command": "uvx",
      "args": ["kedro-mcp@latest"],
      "env": { "FASTMCP_LOG_LEVEL": "ERROR" },
      "disabled": false
    }
  }
}

Now open your assistant and ask:

• “Call kedro_general_instructions and summarise the key conventions.”
• “Call notebook_to_kedro, propose a short plan for my notebook. Wait for APPROVED.”

Tip: @latest auto-refreshes when a newer version is released. Pin a version to freeze behaviour.

---

Features

• Prompt

  • convert_notebook – a short, plain-English prompt that tells the assistant to:
    1. load both guidance docs via tools,
    2. return a brief conversion plan and wait for APPROVED,
    3. then create the Kedro project (venv, install Kedro first), without running pipelines or moving data.

• Tools (read-only)

  • kedro_general_instructions – returns concise conventions for consistent Kedro usage (naming, datasets, parameters, environments, checklists).
  • notebook_to_kedro – returns focused guidance for notebook → Kedro conversion (plan → approval → build) with small templates/diffs.

Both tools just return Markdown text for the assistant to use as context; they do not read your files or send project data anywhere.

---

Installation

Option A — Zero-install (recommended)

Nothing to install into your environment. Your MCP client will run the server locally via uvx using the config above. First run is cached; later runs reuse the cache and only update when you use @latest or change the version.

Option B — From PyPI (when available)

uv pip install kedro-mcp

Then point your client to the console script:

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

Option C — From source (local development)

git clone https://github.com/kedro-org/kedro-mcp.git
cd kedro-mcp
uv pip install -e . --group dev

Config for local path:

{
  "mcpServers": {
    "kedro": {
      "command": "uv",
      "args": ["tool","run","--from",".","kedro-mcp"]
    }
  }
}

---

Manual server test

# Runs the stdio entrypoint (the client usually does this for you)
kedro-mcp

# Or with uvx directly (no install)
uvx kedro-mcp@latest

You should see a small “starting on stdio…” log; then your client will communicate over stdio.

---

Using with assistants

• VS Code – Copilot Chat: once configured, the server and tools appear automatically; you can invoke them by name in chat.
• Claude Desktop / Cursor: add the generic MCP config above; the tools can be called by name (e.g., notebook_to_kedro).

Suggested prompts:

• “Call kedro_general_instructions, then summarise conventions that are missing in this repo.”
• “Call notebook_to_kedro, propose a conversion plan for analysis.ipynb. Wait for APPROVED before creating anything.”

---

Development

# Install dev deps
uv pip install -e . --group dev

# Lint & type-check
ruff check .
mypy src/

Releasing

python -m pip install build twine
python -m build
twine upload dist/*

---

Troubleshooting

• Server not starting: ensure Python 3.10+ and uv are installed; confirm the MCP config points to uvx kedro-mcp@latest or to the kedro-mcp console script.
• Tools don’t appear: restart the assistant; verify the MCP config key matches "kedro" and the client supports stdio servers.
• Version drift: pin a version instead of @latest.

---

License

This project is licensed under the Apache Software License 2.0. See LICENSE.txt for details.

---

Support

• Report issues: https://github.com/kedro-org/kedro-mcp/issues
• MCP specification: https://modelcontextprotocol.io/