Skill Management MCP Server

Instructions

# data-processor:json_fetcher.py
# /// script
# dependencies = ["requests>=2.31.0"]
# ///
import requests
def fetch_json(url): return requests.get(url).json()

{
  "code": "from json_fetcher import fetch_json\ndata = fetch_json('https://api.example.com')\nprint(data)",
  "skill_references": ["data-processor:json_fetcher.py"]
}

{
  "code": "from math_utils import add, multiply\nprint(add(10, 20))",
  "skill_references": ["calculator:math_utils.py"]
}

{
  "code": "from math_utils import add\nfrom stats_utils import mean\nfrom converters import celsius_to_fahrenheit\n\nresult = add(10, 20)\navg = mean([10, 20, 30])\ntemp = celsius_to_fahrenheit(25)\nprint(f'Sum: {result}, Avg: {avg}, Temp: {temp}F')",
  "skill_references": ["calculator:math_utils.py", "calculator:stats_utils.py", "calculator:converters.py"]
}

{
  "code": "from advanced.calculus import derivative_at_point\ndef f(x): return x**2\nprint(derivative_at_point(f, 5))",
  "skill_references": ["calculator:advanced/calculus.py"]
}

{
  "code": "from api_client import fetch_weather\ndata = fetch_weather('London')\nprint(data)",
  "skill_references": ["weather:api_client.py"]
}

{
  "code": "# /// script\n# dependencies = [\n#   \"requests>=2.31.0\",\n#   \"pandas\",\n# ]\n# ///\n\nimport requests\nimport pandas as pd\n\nresponse = requests.get('https://api.example.com/data')\ndf = pd.DataFrame(response.json())\nprint(df.head())"
}

Implementation Reference

• src/skill_mcp/tools/script_tools.py:17-128 (registration)

  Registration of the 'execute_python_code' MCP tool within ScriptTools.get_script_tools(), specifying name, comprehensive description, and inputSchema from ExecutePythonCodeInput.model_json_schema().

              types.Tool(
                  name="execute_python_code",
                  description="""Execute Python code directly without requiring a script file.
  
  RECOMMENDATION: Prefer Python over bash/shell scripts for better portability, error handling, and maintainability.
  
  IMPORTANT: Use this tool instead of creating temporary script files when you need to run quick Python code.
  
  ✅ SUPPORTS PEP 723 INLINE DEPENDENCIES - just like run_skill_script!
  
  FEATURES:
  - **PEP 723 inline dependencies**: Include dependencies directly in code using /// script comments (auto-detected and installed)
  - **Dependency aggregation**: When importing from skills, their PEP 723 dependencies are automatically merged into your code
  - Skill file imports: Reference files from skills using namespace format (skill_name:path/to/file.py)
  - Automatic dependency installation: Code with PEP 723 metadata is run with 'uv run'
  - Environment variable loading: Automatically loads .env files from all referenced skills
  - Clean execution: Temporary file is automatically cleaned up after execution
  
  PARAMETERS:
  - code: Python code to execute (can include PEP 723 dependencies)
  - skill_references: Optional list of skill files to make available for import
                     Format: ["calculator:utils.py", "weather:api/client.py"]
                     The skill directories will be added to PYTHONPATH
                     Environment variables from each skill's .env file will be loaded
  - timeout: Optional timeout in seconds (defaults to 30 seconds if not specified)
  
  CROSS-SKILL IMPORTS - BUILD REUSABLE LIBRARIES:
  Create utility skills once, import them anywhere! Perfect for:
  - Math/statistics libraries (calculator:stats.py)
  - API clients (weather:api_client.py)
  - Data processors (etl:transformers.py)
  - Common utilities (helpers:string_utils.py)
  
  AUTOMATIC DEPENDENCY AGGREGATION:
  When you reference skill files, their PEP 723 dependencies are automatically collected
  and merged into your code! You don't need to redeclare dependencies - just reference
  the modules and their deps are included automatically.
  
  Example - library module with deps:
  ```python
  # data-processor:json_fetcher.py
  # /// script
  # dependencies = ["requests>=2.31.0"]
  # ///
  import requests
  def fetch_json(url): return requests.get(url).json()
  ```
  
  Your code - NO need to declare requests!
  ```json
  {
    "code": "from json_fetcher import fetch_json\\ndata = fetch_json('https://api.example.com')\\nprint(data)",
    "skill_references": ["data-processor:json_fetcher.py"]
  }
  ```
  Dependencies from json_fetcher.py are automatically aggregated!
  
  Import from single skill:
  ```json
  {
    "code": "from math_utils import add, multiply\\nprint(add(10, 20))",
    "skill_references": ["calculator:math_utils.py"]
  }
  ```
  
  Import from multiple skills:
  ```json
  {
    "code": "from math_utils import add\\nfrom stats_utils import mean\\nfrom converters import celsius_to_fahrenheit\\n\\nresult = add(10, 20)\\navg = mean([10, 20, 30])\\ntemp = celsius_to_fahrenheit(25)\\nprint(f'Sum: {result}, Avg: {avg}, Temp: {temp}F')",
    "skill_references": ["calculator:math_utils.py", "calculator:stats_utils.py", "calculator:converters.py"]
  }
  ```
  
  Import from subdirectories:
  ```json
  {
    "code": "from advanced.calculus import derivative_at_point\\ndef f(x): return x**2\\nprint(derivative_at_point(f, 5))",
    "skill_references": ["calculator:advanced/calculus.py"]
  }
  ```
  
  ENVIRONMENT VARIABLES FROM REFERENCED SKILLS:
  When you import from a skill, its environment variables are automatically loaded:
  ```json
  {
    "code": "from api_client import fetch_weather\\ndata = fetch_weather('London')\\nprint(data)",
    "skill_references": ["weather:api_client.py"]
  }
  ```
  If weather:api_client.py uses API_KEY from its .env file, it will be available automatically!
  
  EXAMPLE WITH PEP 723 DEPENDENCIES:
  ```json
  {
    "code": "# /// script\\n# dependencies = [\\n#   \\"requests>=2.31.0\\",\\n#   \\"pandas\\",\\n# ]\\n# ///\\n\\nimport requests\\nimport pandas as pd\\n\\nresponse = requests.get('https://api.example.com/data')\\ndf = pd.DataFrame(response.json())\\nprint(df.head())"
  }
  ```
  
  WHY PYTHON OVER BASH/JS:
  - Better error handling and debugging
  - Rich standard library
  - Cross-platform compatibility
  - Easier to read and maintain
  - Strong typing support
  - Better dependency management
  
  RETURNS: Execution result with:
  - Exit code (0 = success, non-zero = failure)
  - STDOUT (standard output)
  - STDERR (error output)""",
                  inputSchema=ExecutePythonCodeInput.model_json_schema(),
              ),

• src/skill_mcp/tools/script_tools.py:239-267 (handler)

  Primary MCP handler for 'execute_python_code': receives validated input, calls ScriptService for execution, formats stdout/stderr/exit_code into TextContent response.

  async def execute_python_code(
      input_data: ExecutePythonCodeInput,
  ) -> list[types.TextContent]:
      """Execute Python code directly."""
      try:
          result = await ScriptService.execute_python_code(
              input_data.code,
              input_data.skill_references,
              input_data.timeout,
          )
  
          output = "Python Code Execution\n"
          output += f"Exit code: {result.exit_code}\n\n"
  
          if result.stdout:
              output += f"STDOUT:\n{result.stdout}\n"
  
          if result.stderr:
              output += f"STDERR:\n{result.stderr}\n"
  
          if not result.stdout and not result.stderr:
              output += "(No output)\n"
  
          return [types.TextContent(type="text", text=output)]
      except SkillMCPException as e:
          return [types.TextContent(type="text", text=f"Error: {str(e)}")]
      except Exception as e:
          return [types.TextContent(type="text", text=f"Error executing code: {str(e)}")]

• src/skill_mcp/services/script_service.py:285-418 (handler)

  Core execution logic for 'execute_python_code': parses skill_references, aggregates and merges PEP 723 dependencies, creates temp file, executes via 'uv run' or python with proper env/PYTHONPATH, handles timeout and output truncation.

  async def execute_python_code(
      code: str,
      skill_references: Optional[List[str]] = None,
      timeout: Optional[int] = None,
  ) -> ScriptResult:
      """
      Execute Python code directly without requiring a script file.
  
      Supports PEP 723 inline dependencies using /// script comments.
      Can reference skill files using namespace format (skill_name:path/to/file.py).
      Automatically aggregates PEP 723 dependencies from referenced skill files.
  
      Args:
          code: Python code to execute (can include PEP 723 dependencies)
          skill_references: List of skill files in namespace format to make available
                          e.g., ["calculator:utils.py", "weather:api/client.py"]
          timeout: Optional timeout in seconds (defaults to SCRIPT_TIMEOUT_SECONDS if not specified)
  
      Returns:
          ScriptResult with stdout, stderr, and exit code
  
      Raises:
          ScriptExecutionError: If execution fails
      """
      # Use provided timeout or fall back to default
      script_timeout = timeout if timeout is not None else SCRIPT_TIMEOUT_SECONDS
  
      temp_file = None
      try:
          # Parse skill references and collect dependencies
          env = os.environ.copy()
          python_paths: List[str] = []
          aggregated_deps: List[str] = []
          processed_skills: set[str] = set()  # Track which skills we've processed
  
          if skill_references:
              for ref in skill_references:
                  # Parse namespace format: skill_name:path/to/file.py
                  if ":" not in ref:
                      raise ScriptExecutionError(
                          f"Invalid skill reference format: '{ref}'. Expected 'skill_name:path/to/file.py'"
                      )
  
                  skill_name, file_path = ref.split(":", 1)
                  skill_dir = SKILLS_DIR / skill_name
  
                  if not skill_dir.exists():
                      raise SkillNotFoundError(f"Skill '{skill_name}' does not exist")
  
                  # Add skill directory to PYTHONPATH
                  if str(skill_dir) not in python_paths:
                      python_paths.append(str(skill_dir))
  
                  # Load environment variables from this skill (only once per skill)
                  if skill_name not in processed_skills:
                      try:
                          skill_env = EnvironmentService.load_skill_env(skill_name)
                          env.update(skill_env)
                          processed_skills.add(skill_name)
                      except Exception:
                          # If we can't load env vars, just skip (skill may not have .env file)
                          processed_skills.add(skill_name)
  
                  # Read the referenced file and extract its dependencies
                  ref_file_path = skill_dir / file_path
                  if ref_file_path.exists() and ref_file_path.is_file():
                      try:
                          ref_content = ref_file_path.read_text(encoding="utf-8")
                          ref_deps = extract_pep723_dependencies(ref_content)
                          aggregated_deps.extend(ref_deps)
                      except Exception:
                          # If we can't read or parse the file, just skip dependency extraction
                          pass
  
          # Merge aggregated dependencies into the code
          if aggregated_deps:
              code = merge_dependencies(code, aggregated_deps)
  
          # Create temporary Python file with merged code
          with tempfile.NamedTemporaryFile(
              mode="w", suffix=".py", delete=False, encoding="utf-8"
          ) as f:
              f.write(code)
              temp_file = Path(f.name)
  
          # Add paths to PYTHONPATH
          if python_paths:
              existing_path = env.get("PYTHONPATH", "")
              new_paths = ":".join(python_paths)
              env["PYTHONPATH"] = f"{new_paths}:{existing_path}" if existing_path else new_paths
  
          # Check if code has PEP 723 dependencies (same logic as has_uv_dependencies)
          has_deps = "# /// script" in code or "# /// pyproject" in code
  
          # Build command
          if has_deps:
              cmd = ["uv", "run", str(temp_file)]
          else:
              cmd = [DEFAULT_PYTHON_INTERPRETER, str(temp_file)]
  
          # Execute
          result = subprocess.run(
              cmd,
              capture_output=True,
              text=True,
              env=env,
              timeout=script_timeout,
          )
  
          # Truncate output if needed
          stdout = result.stdout
          if len(stdout) > MAX_OUTPUT_SIZE:
              stdout = stdout[:MAX_OUTPUT_SIZE] + "\n... (output truncated)"
  
          stderr = result.stderr
          if len(stderr) > MAX_OUTPUT_SIZE:
              stderr = stderr[:MAX_OUTPUT_SIZE] + "\n... (output truncated)"
  
          return ScriptResult(result.returncode, stdout, stderr)
  
      except subprocess.TimeoutExpired:
          raise ScriptExecutionError(f"Code execution timed out ({script_timeout} seconds)")
      except (SkillNotFoundError, ScriptExecutionError):
          raise
      except Exception as e:
          raise ScriptExecutionError(f"Failed to execute code: {str(e)}")
      finally:
          # Clean up temporary file
          if temp_file and temp_file.exists():
              try:
                  temp_file.unlink()
              except Exception:
                  pass

• src/skill_mcp/models.py:73-85 (schema)

  Pydantic BaseModel defining input schema for 'execute_python_code' tool: required 'code' str, optional 'skill_references' list and 'timeout' int.

  class ExecutePythonCodeInput(BaseModel):
      """Input for executing Python code directly."""
  
      code: str = Field(description="Python code to execute (can include PEP 723 dependencies)")
      skill_references: Optional[List[str]] = Field(
          default=None,
          description="Optional list of skill files to import using namespace format (e.g., 'calculator:utils.py')",
      )
      timeout: Optional[int] = Field(
          default=None,
          description="Optional timeout in seconds (defaults to 30 seconds if not specified)",
      )

• src/skill_mcp/services/script_service.py:22-54 (helper)

  Helper function to parse and extract PEP 723 dependencies from Python code or referenced skill files for automatic aggregation.

  def extract_pep723_dependencies(content: str) -> List[str]:
      """
      Extract dependencies from PEP 723 metadata in code.
  
      Args:
          content: Python code or file content
  
      Returns:
          List of dependency strings (e.g., ["requests>=2.31.0", "pandas"])
      """
      # Match PEP 723 script block
      pattern = r"#\s*///\s*script\s*\n(.*?)#\s*///\s*$"
      match = re.search(pattern, content, re.MULTILINE | re.DOTALL)
  
      if not match:
          return []
  
      metadata_block = match.group(1)
  
      # Extract dependencies array
      deps_pattern = r"dependencies\s*=\s*\[(.*?)\]"
      deps_match = re.search(deps_pattern, metadata_block, re.DOTALL)
  
      if not deps_match:
          return []
  
      deps_content = deps_match.group(1)
  
      # Extract individual dependency strings
      dep_strings = re.findall(r'["\']([^"\']+)["\']', deps_content)
  
      return dep_strings