feat: add MCP (Model Context Protocol) server

src/mcp/__init__.py

@@ -0,0 +1,14 @@
+"""MCP Server"""
+
+
+def create_mcp_server():
+    from .server import mcp
+    return mcp
+
+
+def run_server():
+    from .server import run_server as _run
+    _run()
+
+
+__all__ = ["create_mcp_server", "run_server"]

src/mcp/server.py

@@ -0,0 +1,19 @@
+"""MCP Server - FastMCP implementation"""
+
+import logging
+from fastmcp import FastMCP
+
+logger = logging.getLogger("codeatlas.mcp")
+
+mcp = FastMCP("CodeAtlas")
+
+from . import tools  # noqa: F401, E402
+
+
+def run_server():
+    logger.info("Starting CodeAtlas MCP Server...")
+    mcp.run()
+
+
+if __name__ == "__main__":
+    run_server()

src/mcp/tools.py

@@ -0,0 +1,99 @@
+"""MCP Tools - CodeAtlas functions exposed via MCP protocol"""
+
+import logging
+from typing import Optional
+
+from .server import mcp
+from ..core.repository import RepositoryLoader
+from ..core.analyzer import CodeAnalyzer
+from ..core.diagram import DiagramGenerator
+
+logger = logging.getLogger("codeatlas.mcp.tools")
+
+
+@mcp.tool()
+def analyze_codebase(api_key: str, github_url: Optional[str] = None, file_path: Optional[str] = None, model_name: str = "gemini-2.5-flash") -> str:
+    """Analyze a codebase and generate a Graphviz DOT architecture diagram."""
+    logger.info(f"analyze_codebase: url={github_url}, file={file_path}, model={model_name}")
+    
+    if not github_url and not file_path:
+        return "Error: Please provide either a GitHub URL or a file path."
+    if not api_key:
+        return "Error: API key is required."
+    
+    loader = RepositoryLoader()
+    result = loader.load_from_file(file_path) if file_path else loader.load_from_github(github_url)
+    
+    if result.error:
+        return f"Error: {result.error}"
+    
+    logger.info(f"Loaded: {result.stats.files_processed} files, {result.stats.total_characters:,} chars")
+    
+    analyzer = CodeAnalyzer(api_key=api_key, model_name=model_name)
+    analysis = analyzer.generate_diagram(result.context)
+    
+    if not analysis.success:
+        return f"Error: {analysis.error}"
+    
+    DiagramGenerator()._save_diagram(analysis.content, "raw", result.repo_name)
+    return analysis.content
+
+
+@mcp.tool()
+def get_architecture_summary(api_key: str, github_url: Optional[str] = None, file_path: Optional[str] = None, model_name: str = "gemini-2.5-flash") -> str:
+    """Generate a text summary of a codebase's architecture."""
+    logger.info(f"get_architecture_summary: url={github_url}, file={file_path}")
+    
+    if not github_url and not file_path:
+        return "Error: Please provide either a GitHub URL or a file path."
+    if not api_key:
+        return "Error: API key is required."
+    
+    loader = RepositoryLoader()
+    result = loader.load_from_file(file_path) if file_path else loader.load_from_github(github_url)
+    
+    if result.error:
+        return f"Error: {result.error}"
+    
+    analyzer = CodeAnalyzer(api_key=api_key, model_name=model_name)
+    analysis = analyzer.generate_summary(result.context)
+    
+    return analysis.content if analysis.success else f"Error: {analysis.error}"
+
+
+@mcp.tool()
+def chat_with_codebase(api_key: str, question: str, github_url: Optional[str] = None, file_path: Optional[str] = None, model_name: str = "gemini-2.5-flash") -> str:
+    """Ask questions about a codebase and get AI-powered answers."""
+    logger.info(f"chat_with_codebase: question={question[:50]}...")
+    
+    if not github_url and not file_path:
+        return "Error: Please provide either a GitHub URL or a file path."
+    if not api_key:
+        return "Error: API key is required."
+    if not question or not question.strip():
+        return "Error: Please provide a question."
+    
+    loader = RepositoryLoader()
+    result = loader.load_from_file(file_path) if file_path else loader.load_from_github(github_url)
+    
+    if result.error:
+        return f"Error: {result.error}"
+    
+    analyzer = CodeAnalyzer(api_key=api_key, model_name=model_name)
+    response = analyzer.chat(question, result.context)
+    
+    return response.content if response.success else f"Error: {response.error}"
+
+
+@mcp.tool()
+def list_recent_analyses() -> str:
+    """List recently analyzed codebases."""
+    history = DiagramGenerator().get_history(limit=10)
+    
+    if not history:
+        return "No recent analyses found."
+    
+    lines = ["Recent Analyses:"]
+    for i, d in enumerate(history, 1):
+        lines.append(f"{i}. {d.repo_name} — {d.formatted_timestamp}")
+    return "\n".join(lines)