MCP Local Codebase Search

This project is a simple MCP (Model Context Protocol) server for local Python codebase search.

It provides tools to:

• index Python files into a SQLite database,

• search symbols (functions/classes),

• find files importing specific modules,

• read file contents.

Features

• index_project()

  • Scans the current working directory and indexes .py files.

  • Stores symbol and dependency data in db/code_index.db.

• find_symbol(name)

  • Searches function/class names using LIKE matching.

  • Returns symbol name, type, file path, and line number.

• find_import(module)

  • Finds files that import a specific module.

• read_file(path)

  • Reads and returns full file content with encoding fallback (utf8, latin-1).

Project Structure

• server.py: MCP server entrypoint and tool definitions.

• indexer.py: Indexing logic to SQLite.

• parser.py: Symbol extraction (FunctionDef, ClassDef) using ast.

• dependency.py: Import extraction using ast.

• search.py: Search queries against the database.

• db/: Index database location (code_index.db).

Prerequisites

• Python 3.10+ (recommended)

• pip

Installation

1. Clone this repository.

2. Go to the project folder.

3. Create a virtual environment.

4. Activate the virtual environment.

5. Install dependencies.

Windows PowerShell example:

python -m venv venv
.\venv\Scripts\Activate.ps1
pip install -r requirements.txt

Run Locally

Start the MCP server:

python server.py

Available tools:

• index_project

• find_symbol

• find_import

• read_file

MCP Configuration

Add this to your MCP client configuration so it runs as an MCP server:

{
  "mcpServers": {
    "code-intelligence": {
      "command": "python",
      "args": [
        "path-to-your-server"
      ]
    }
  }
}

Recommended Usage Flow

1. Run index_project once to build the initial index.

2. Use find_symbol to locate functions/classes.

3. Use find_import to trace module usage.

4. Re-run index_project after major code changes.

Notes

• Non-Python files are not indexed.

• Ignored directories during indexing: .git, node_modules, __pycache__, dist, build.

• The index database is local and should not be committed (already covered by .gitignore).