Add inline python extension

[DOsinga]

This makes it possible to have python code inside of a recipe

[DOsinga]

This should work:

version: "1.0.0"
title: "Word Counter"
description: "A recipe that counts words in text using an inline Python extension"

instructions: |
  This recipe provides a simple word counting tool. You can:
  1. Count words in a text string
  2. Get statistics about the text (unique words, average word length)
  3. Generate word clouds from the text

parameters:
  - key: text
    input_type: string
    requirement: required
    description: "The text to analyze"

extensions:
  - type: inline_python
    name: word_counter
    code: |
      from mcp.server.fastmcp import FastMCP
      import json
      from collections import Counter
      import numpy as np
      from wordcloud import WordCloud
      import matplotlib.pyplot as plt
      import base64
      from io import BytesIO
      
      mcp = FastMCP("word_counter")
      
      @mcp.tool()
      def count_words(text: str) -> str:
          """Count the number of words in a text string and generate statistics"""
          words = text.split()
          word_count = len(words)
          unique_words = len(set(words))
          avg_length = sum(len(word) for word in words) / word_count if word_count > 0 else 0
          
          # Generate word cloud
          wordcloud = WordCloud(width=800, height=400, background_color='white').generate(text)
          
          # Save word cloud to base64
          plt.figure(figsize=(10, 5))
          plt.imshow(wordcloud, interpolation='bilinear')
          plt.axis('off')
          
          # Save to bytes
          buf = BytesIO()
          plt.savefig(buf, format='png', bbox_inches='tight', pad_inches=0)
          buf.seek(0)
          img_base64 = base64.b64encode(buf.getvalue()).decode()
          plt.close()
          
          result = {
              "total_words": word_count,
              "unique_words": unique_words,
              "average_word_length": round(avg_length, 2),
              "most_common": dict(Counter(words).most_common(5)),
              "wordcloud_base64": img_base64
          }
          
          return json.dumps(result, indent=2)
          
      if __name__ == "__main__":
          mcp.run()
    timeout: 300
    description: "Count words and provide text statistics with visualization"
    dependencies:
      - "mcp"
      - "numpy"
      - "matplotlib"
      - "wordcloud"
      - "beautifulsoup4"
      - "html2text"
      - "requests"

prompt: |
  You are a helpful assistant that can analyze text using word counting tools.
  The user has provided the following text to analyze: {{ text }}
  
  Use the word_counter__count_words tool to analyze this text and provide insights about:
  - Total word count
  - Number of unique words
  - Average word length
  - Most commonly used words
  - A word cloud visualization of the text

[michaelneale]

diabolically good - I love it. And security reasons why this is a bad idea?

[DOsinga]

And security reasons why this is a bad idea?

I think this can actually pave the ways to make things more secure - if you have a recipe that refers to an MCP server that is unsigned (as they all are today) and then you accept the recipe, the author of the MCP server could change the implementation and do bad things. here are least you see what is actually happening. I don't think it makes recipes less save, but very save they are not to begin with

[DOsinga]

assigned to you @michaelneale

[michaelneale]

updated this to main and had to make some other changes for new typechecks and openapi.

@DOsinga wasn't able to run that example, having trouble speaking uv to know if this means I really need a different py installed:

Warning: Failed to start extension 'word_counter': Failed to start the MCP server from configuration `InlinePython(word_counter: 1341 chars)` `Call to '' failed for 'initialize'. Error from mcp-server: Stdio process error:   × No solution found when resolving tool dependencies:
  ╰─▶ Because python was not found in the package registry and you require
      python, we can conclude that your requirements are unsatisfiable.
`
Continuing without extension 'word_counter'

ie it was trying to do python - but failing I guess once it is bootstrapping from the python-mcp side of things?

[DOsinga]

updated this to main and had to make some other changes for new typechecks and openapi.

@DOsinga wasn't able to run that example, having trouble speaking uv to know if this means I really need a different py installed:

Warning: Failed to start extension 'word_counter': Failed to start the MCP server from configuration `InlinePython(word_counter: 1341 chars)` `Call to '' failed for 'initialize'. Error from mcp-server: Stdio process error:   × No solution found when resolving tool dependencies:
  ╰─▶ Because python was not found in the package registry and you require
      python, we can conclude that your requirements are unsatisfiable.
`
Continuing without extension 'word_counter'

ie it was trying to do python - but failing I guess once it is bootstrapping from the python-mcp side of things?

was this with Warp on or off? we should probably do better with MCP error handling here (maybe the new version will anyway)

[michaelneale]

@DOsinga with it on - with it off I got another error which was clearer (usually I can run uv things - didn't quite follow this error). Am ok with merging this though as I think it should work - just not sure how reliable in the face of dependencies/registries (but small risk)