token-efficient-dev

Token-Efficient Development Workflow

Master the art of professional AI-assisted development by combining Code Contractor MCP tools with Cursor's native capabilities for maximum efficiency.

Core Philosophy

Token cost hierarchy (cheapest to most expensive):

1. Structure only - get_file_outline (~50 tokens)
2. Targeted extraction - extract_code_element (~200 tokens)
3. Surgical edits - AST tools, insert_at_line (~100 tokens)
4. Search operations - search_code, Grep (~300 tokens)
5. Full file read - Cursor Read (~1000+ tokens per 100 lines)
6. String replace - Cursor StrReplace (~2x content tokens)

Reading Code - Priority Order

Step 1: Always Start with Outline

get_file_outline(path: "src/server.js")

Returns: function names, classes, line numbers (~50 tokens) Never read a file without outlining first!

Step 2: Extract Only What You Need

extract_code_element(
  path: "src/server.js",
  element_name: "handleRequest",
  type: "function"
)

Returns: just that function with context (~200 tokens)

Step 3: Use Cursor Read Only When Necessary

• Small files (<100 lines)
• Need to see imports/structure
• Config files (package.json, etc.)

Writing Code - Priority Order

Priority 1: AST Replace (Best)

ast_replace_element(
  path: "src/auth.js",
  element_name: "validateToken",
  element_type: "function",
  new_content: "async function validateToken(token) { ... }"
)

Why best: Only sends new code, finds function by name (not line numbers)

Priority 2: Insert Operations

insert_at_line(path, line_number, content)
append_to_file(path, content)
prepend_to_file(path, content)
ast_add_import(path, module_source, named_imports)

Why good: Only sends new content

Priority 3: Marker-Based

insert_relative_to_marker(path, marker: "// API Routes", position: "after", content)
replace_between_markers(path, start_marker, end_marker, content)

Why decent: Short marker + new content