MCP server for codebase indexing and semantic search.

uv add mcp httpx fastapi "uvicorn[standard]" toml websockets
uv sync

The configuration file is automatically created at ~/.acemcp/settings.toml on first run with default values.

Edit ~/.acemcp/settings.toml to configure:

BATCH_SIZE = 10
MAX_LINES_PER_BLOB = 800
BASE_URL = "https://your-api-endpoint.com"
TOKEN = "your-bearer-token-here"
TEXT_EXTENSIONS = [".py", ".js", ".ts", ...]
EXCLUDE_PATTERNS = [".venv", "node_modules", ".git", "__pycache__", "*.pyc", ...]

Configuration options:

• BATCH_SIZE: Number of files to upload per batch (default: 10)
• MAX_LINES_PER_BLOB: Maximum lines per blob before splitting large files (default: 800)
• BASE_URL: API endpoint URL
• TOKEN: Authentication token
• TEXT_EXTENSIONS: List of file extensions to index
• EXCLUDE_PATTERNS: List of patterns to exclude from indexing (supports wildcards like *.pyc)

You can also configure via:

• Command line arguments (highest priority): --base-url, --token
• Web management interface (updates user config file)
• Environment variables with ACEMCP_ prefix

Add the following to your MCP client configuration (e.g., Claude Desktop):

{
  "mcpServers": {
    "acemcp": {
      "command": "uvx",
      "args": [
        "acemcp"
      ]
    }
  }
}

Available command line arguments:

• --base-url: Override BASE_URL configuration
• --token: Override TOKEN configuration
• --web-port: Enable web management interface on specified port (e.g., 8080)

To enable the web management interface, add the --web-port argument:

{
  "mcpServers": {
    "acemcp": {
      "command": "uvx",
      "args": [
        "acemcp",
        "--web-port",
        "8888"
      ]
    }
  }
}

Then access the management interface at http://localhost:8888

Web Management Features:

• Configuration Management: View and edit server configuration (BASE_URL, TOKEN, BATCH_SIZE, MAX_LINES_PER_BLOB, TEXT_EXTENSIONS)
• Real-time Logs: Monitor server logs in real-time via WebSocket connection
• Tool Debugger: Test and debug MCP tools directly from the web interface
  • Test index_code tool with any project path
  • Test search_context tool with project path and query
  • View formatted results and error messages

Search for relevant code context based on a query. This tool automatically performs incremental indexing before searching, ensuring results are always up-to-date. It performs semantic search across your codebase and returns formatted text snippets showing where relevant code is located.

Key Features:

• Automatic Incremental Indexing: Before each search, the tool automatically indexes only new or modified files, skipping unchanged files for efficiency
• No Manual Indexing Required: You don't need to manually index your project - just search and the tool handles indexing automatically
• Always Up-to-Date: Search results reflect the current state of your codebase

Parameters:

• project_root_path (string): Absolute path to the project root directory
  • IMPORTANT: Use forward slashes (/) as path separators, even on Windows
  • Windows example: C:/Users/username/projects/myproject
  • Linux/Mac example: /home/username/projects/myproject
• query (string): Natural language search query to find relevant code context
  • Use descriptive keywords related to what you're looking for
  • The tool performs semantic matching, not just keyword search
  • Returns code snippets with file paths and line numbers

What it returns:

• Formatted text snippets from files that match your query
• File paths and line numbers for each snippet
• Context around the relevant code sections
• Multiple results ranked by relevance

Query Examples:

1. Finding configuration code:

   {
     "project_root_path": "C:/Users/username/projects/myproject",
     "query": "查找所有调用 get_model 的地方"
   }

   Returns: Code related to logging setup, logger initialization, and configuration

2. Finding authentication logic:

   {
     "project_root_path": "C:/Users/username/projects/myproject",
     "query": "user authentication login password validation"
   }

   Returns: Authentication handlers, login functions, password validation code

3. Finding database code:

   {
     "project_root_path": "C:/Users/username/projects/myproject",
     "query": "database connection pool initialization"
   }

   Returns: Database connection setup, pool configuration, initialization code

4. Finding error handling:

   {
     "project_root_path": "C:/Users/username/projects/myproject",
     "query": "error handling exception try catch"
   }

   Returns: Error handling patterns, exception handlers, try-catch blocks

5. Finding API endpoints:

   {
     "project_root_path": "C:/Users/username/projects/myproject",
     "query": "API endpoint routes HTTP handlers"
   }

   Returns: API route definitions, HTTP handlers, endpoint implementations

Tips for better results:

• Use multiple related keywords (e.g., "logging configuration setup" instead of just "logging")
• Include technical terms specific to what you're looking for
• Describe the functionality rather than exact variable names
• Try different phrasings if the first query doesn't return what you need

Indexing Features:

• Incremental Indexing: Only new or modified files are uploaded, unchanged files are skipped
• Hash-based Deduplication: Files are identified by SHA-256 hash of path + content
• Automatic Retry: Network requests are automatically retried up to 3 times with exponential backoff (1s, 2s, 4s)
• Batch Resilience: If a batch upload fails after retries, the tool continues with the next batch
• File Splitting: Large files are automatically split into multiple blobs (default: 800 lines per blob)
• Exclude Patterns: Automatically skips virtual environments, node_modules, .git, build artifacts, etc.

Search Features:

• Automatic Retry: Search requests are automatically retried up to 3 times with exponential backoff (2s, 4s, 8s)
• Graceful Degradation: Returns a clear error message if the search fails after all retries
• Timeout Handling: Uses a 60-second timeout to handle long-running searches
• Empty Result Handling: Returns a helpful message if no relevant code is found

Default Exclude Patterns:

.venv, venv, .env, env, node_modules, .git, .svn, .hg, __pycache__,
.pytest_cache, .mypy_cache, .tox, .eggs, *.egg-info, dist, build,
.idea, .vscode, .DS_Store, *.pyc, *.pyo, *.pyd, .Python,
pip-log.txt, pip-delete-this-directory.txt, .coverage, htmlcov,
.gradle, target, bin, obj

Patterns support wildcards (*, ?) and match against directory/file names or paths.

1. Start the MCP server (automatically started by MCP client)
2. Use search_context to search for code context
   • The tool automatically indexes your project before searching
   • Incremental indexing ensures only new/modified files are uploaded
   • No manual indexing step required!

• Configuration: ~/.acemcp/settings.toml
• Indexed projects: ~/.acemcp/data/projects.json (fixed location)
• Log files: ~/.acemcp/log/acemcp.log (with automatic rotation)
• Projects are identified by their absolute path (normalized with forward slashes)

The application automatically logs to ~/.acemcp/log/acemcp.log with the following features:

• Console output: INFO level and above (colored output)
• File output: DEBUG level and above (detailed format with module, function, and line number)
• Automatic rotation: Log files are rotated when they reach 5MB
• Retention: Maximum of 10 log files are kept
• Compression: Rotated log files are automatically compressed to .zip format
• Thread-safe: Logging is thread-safe for concurrent operations

Log format:

2025-11-06 13:51:25 | INFO     | acemcp.server:main:103 - Starting acemcp MCP server...

The log files are automatically created on first run and require no manual configuration.

The web management interface provides:

• Real-time server status monitoring
• Live log streaming via WebSocket
• Configuration viewing (current settings)
• Project statistics (number of indexed projects)

To enable the web interface, use the --web-port argument when starting the server.

Features:

• Real-time log display with auto-scroll
• Server status and metrics
• Configuration overview
• Responsive design with Tailwind CSS
• No build step required (uses CDN resources)