tokennuke 0.2.0

Nuke your token usage. Intelligent code indexer with O(1) symbol retrieval, hybrid search, and call graph analysis.

TokenNuke

Nuke your token usage. Intelligent code indexer MCP server with O(1) symbol retrieval, hybrid search, and call graph analysis.

TokenNuke indexes your codebase using tree-sitter, extracts every function, class, method, constant, and type with byte offsets, then serves them to AI agents via MCP. Instead of reading entire files, agents retrieve exactly the symbols they need — saving 80-95% of tokens.

Features

• O(1) Symbol Retrieval — Byte-offset reads. No file scanning.
• 10 Languages — Python, JavaScript, TypeScript, Go, Rust, Java, C, C++, C#, Ruby
• Hybrid Search — FTS5 keyword search + vector embeddings via Reciprocal Rank Fusion
• Call Graph — Trace "who calls this?" and "what does this call?" with depth traversal
• Incremental Indexing — SHA-256 per file. Only re-indexes what changed.
• Full-Text Search — Search raw file contents (strings, comments, config values)
• No File Limit — SQLite backend, not JSON. Index repos with 100K+ files.
• Dual Transport — stdio (local) or streamable-http (remote)
• Security — Path traversal protection, secret file detection, binary filtering

Install

pip install tokennuke

Quick Start

Claude Code / Claude Desktop

Add to your MCP config (~/.claude.json or claude_desktop_config.json):

{
  "mcpServers": {
    "tokennuke": {
      "command": "tokennuke",
      "args": []
    }
  }
}

CLI

# Start in stdio mode (default)
tokennuke

# Start as HTTP server
tokennuke --transport streamable-http --port 5100

13 MCP Tools

Tool	Description
index_folder	Index a local directory
index_repo	Clone and index a Git repo (GitHub/GitLab/Bitbucket)
list_repos	List all indexed repositories
invalidate_cache	Force full re-index
file_tree	Directory tree with file counts
file_outline	All symbols in a single file
repo_outline	All symbols in repo (summary)
get_symbol	Full source of one symbol (O(1) byte seek)
get_symbols	Batch get multiple symbols
search_symbols	Hybrid FTS5 + vector search
search_text	Full-text search in file contents
get_callees	What does this function call?
get_callers	Who calls this function?

How It Works

1. Index: Tree-sitter parses source files into ASTs. We walk each AST extracting symbols (functions, classes, methods, constants, types) with their exact byte offsets.
2. Store: Symbols go into per-repo SQLite databases with FTS5 indexes and sqlite-vec embeddings.
3. Search: Hybrid search combines BM25 keyword matching with semantic vector similarity via Reciprocal Rank Fusion.
4. Retrieve: get_symbol seeks to the exact byte offset in the source file and reads only that symbol's bytes. No wasted tokens.
5. Graph: Call expressions are extracted from function bodies and stored as edges. Callee names are resolved to symbol IDs after indexing.

Token Savings Example

Method	Tokens	Savings
Read entire file (500 lines)	~4,000	—
file_outline + get_symbol	~200-400	90-95%

Development

git clone https://github.com/BigJai/tokennuke
cd tokennuke
pip install -e ".[dev]"
pytest tests/

License

MIT