mcp-server-collector

Instructions

Implementation Reference

• src/mcp_server_collector/extract.py:8-30 (handler)

  Core handler function that uses OpenAI to extract MCP servers from the provided content using a JSON-structured prompt.

  async def extract_mcp_servers_from_content(content: str) -> str | None:
      client = OpenAI(
          api_key=os.getenv("OPENAI_API_KEY"),
          base_url=os.getenv("OPENAI_BASE_URL"),
      )
  
      user_content = extract_mcp_servers_prompt.format(content=content)
  
      logger.info(f"Extract prompt: {user_content}")
  
      chat_completion = client.chat.completions.create(
          messages=[
              {
                  "role": "user",
                  "content": user_content,
              }
          ],
          model=os.getenv("OPENAI_MODEL"),
          response_format={"type": "json_object"},
      )
  
      return chat_completion.choices[0].message.content

• src/mcp_server_collector/server.py:32-45 (registration)

  Registers the tool in the MCP server's list_tools() method, defining name, description, and input schema.

  types.Tool(
      name="extract-mcp-servers-from-content",
      description="Extract MCP Servers from given content",
      inputSchema={
          "type": "object",
          "properties": {
              "content": {
                  "type": "string",
                  "description": "content containing mcp servers",
              },
          },
          "required": ["content"],
      },
  ),

• src/mcp_server_collector/server.py:35-44 (schema)

  Input schema defining the expected 'content' parameter for the tool.

  inputSchema={
      "type": "object",
      "properties": {
          "content": {
              "type": "string",
              "description": "content containing mcp servers",
          },
      },
      "required": ["content"],
  },

• src/mcp_server_collector/server.py:66-117 (handler)

  MCP server call_tool handler that dispatches to extract logic for this tool and formats the response.

  @server.call_tool()
  async def handle_call_tool(
      name: str, arguments: dict | None
  ) -> list[types.TextContent | types.ImageContent | types.EmbeddedResource]:
      if not arguments:
          raise ValueError("Missing arguments")
  
      content = None
      
      match name:
          case "extract-mcp-servers-from-url":
              url = arguments.get("url")
              if not url:
                  raise ValueError("Missing url")
  
              content = await call_fetch_tool(url)
              
          case "extract-mcp-servers-from-content":
              content = arguments.get("content")
              
          case "submit-mcp-server":
              url = arguments.get("url")
              avatar_url = arguments.get("avatar_url") or ""
              result = await submit_mcp_server(url, avatar_url)
              content = json.dumps(result)
  
              return [
                  types.TextContent(
                      type="text",
                      text=content,
                  )
              ]
          case _:
              raise ValueError(f"Unknown tool: {name}")
  
      if not content:
          raise ValueError("Missing content")
  
      logger.info(f"Fetched content from {url}: {content}")
  
      mcp_servers = await extract_mcp_servers_from_content(content)
      if not mcp_servers:
          raise ValueError("Extracted no MCP Servers")
  
      logger.info(f"Extracted MCP Servers from {url}: {mcp_servers}")
  
      return [
          types.TextContent(
              type="text",
              text=mcp_servers,
          )
      ]   

• src/mcp_server_collector/prompts.py:1-21 (helper)

  Prompt template used in the extraction handler to guide the LLM in parsing MCP server details from content.

  extract_mcp_servers_prompt = """Please extract all MCP Servers from the following content and return a JSON array. Each item should contain:
  - name: extracted from the repository name in the URL
  - title: a human readable title
  - description: a brief description of the server
  - url: the full GitHub repository URL
  - author_name: extracted from the GitHub username in the URL
  
  Example response format:
  [
      {{
          "name": "mcp-server-example",
          "title": "MCP Server Example",
          "description": "A sample MCP server implementation",
          "url": "https://github.com/username/mcp-server-example",
          "author_name": "username"
      }}
  ]
  
  Content to analyze:
  {content}
  """