mcp-toolz

Instructions

Implementation Reference

• src/mcp_server/server.py:549-567 (handler)

  Main handler logic for the 'ask_claude' tool: retrieves context by ID, instantiates ClaudeClient, calls get_second_opinion, optionally stores response, and returns formatted output.

  if name == "ask_claude":
      context_id = arguments["context_id"]
      question = arguments.get("question")
      context = self.storage.get_context(context_id)
      if not context:
          return [TextContent(type="text", text=f"Context {context_id} not found")]
  
      try:
          claude_client = ClaudeClient()
          response = claude_client.get_second_opinion(context, question)
  
          # Only save to database if it's a generic second opinion (no custom question)
          if not question:
              self.storage.update_claude_response(context_id, response)
  
          header = "Claude's Answer:" if question else "Claude's Opinion:"
          return [TextContent(type="text", text=f"{header}\n\n{response}")]
      except ValueError as e:
          return [TextContent(type="text", text=f"Error: {e}")]

• src/mcp_server/server.py:331-347 (schema)

  Tool schema definition including name, description, and input schema requiring 'context_id' and optional 'question'.

  Tool(
      name="ask_claude",
      description="Ask Claude a question about a context entry, or get a general second opinion",
      inputSchema={
          "type": "object",
          "properties": {
              "context_id": {"type": "string", "description": "Context ID to ask about"},
              "question": {
                  "type": "string",
                  "description": (
                      "Optional specific question to ask about the context. If not provided, gets a general second opinion."
                  ),
              },
          },
          "required": ["context_id"],
      },
  ),

• src/context_manager/anthropic_client.py:24-65 (helper)

  Core helper function in ClaudeClient that formats the context prompt and calls the Anthropic Claude API to generate the response.

      def get_second_opinion(self, context: ContextEntry, question: str | None = None) -> str:
          """Get Claude's second opinion on a context, or answer a specific question.
  
          Args:
              context: The context entry to analyze
              question: Optional specific question to ask. If None, provides general second opinion.
          """
          if question:
              # Custom question mode
              system_prompt = """You are a senior software engineering consultant answering questions about code, \
  architecture decisions, and implementation plans.
  
  Provide clear, actionable answers based on the context provided."""
              user_content = self._format_context_for_claude(context, question)
          else:
              # Generic second opinion mode
              system_prompt = """You are a senior software engineering consultant providing second opinions on code, \
  architecture decisions, and implementation plans.
  
  Your role is to:
  - Provide constructive, balanced feedback
  - Highlight both strengths and potential issues
  - Suggest alternatives when appropriate
  - Point out edge cases or security concerns
  - Be concise but thorough
  
  Format your response clearly with sections as needed."""
              user_content = self._format_context_for_claude(context)
  
          response = self.client.messages.create(
              model=self.model,
              max_tokens=4096,
              system=system_prompt,
              messages=[
                  {"role": "user", "content": user_content},
              ],
              temperature=0.7,
          )
  
          if response.content and isinstance(response.content[0], TextBlock):
              return response.content[0].text
          return ""