TokenSqueeze

A codebase indexing and symbol retrieval for Claude Code token optimization.

Index your project once, then let Claude retrieve precise function signatures, class outlines, and full symbol source — without dumping entire files into context.

Install

npx token-squeeze@latest

This installs the Claude Code plugin and downloads the correct binary for your platform. Restart Claude Code after installing.

Manual install

If you prefer to build from source (requires .NET 9 SDK):

git clone https://github.com/mpool/token-squeeze.git
cd token-squeeze
./plugin/build.sh

Then copy the plugin/ contents to ~/.claude/plugins/token-squeeze/.

What it does

TokenSqueeze gives Claude Code two interfaces:

MCP Tools (automatic)

The plugin bundles an MCP server (stdio transport) that Claude calls directly — no slash commands needed:

| Tool | Description | |------|-------------| | read_file_outline | Show symbols in a file (functions, classes, methods, types) | | read_symbol_source | Get the full source of one or more symbols by ID | | search_symbols | Search symbols by name, signature, or docstring |

Skills (slash commands)

| Skill | Description | |-------|-------------| | /token-squeeze:index | Index a directory's symbols (run this first) | | /token-squeeze:explore | Research how code works by tracing implementations | | /token-squeeze:savings | Estimate tokens saved this session |

Supported Languages

Python, JavaScript, TypeScript, C#, C, C++

How it works

Source files  -->  tree-sitter parser  -->  Symbol index (JSON)
                                              |
Claude Code  <--  CLI query commands  <-------+

Symbols are stored in <project>/.cache/ as JSON indexes alongside raw source snapshots. Each project's index lives alongside its source code. Claude queries them through the CLI, getting just the symbols it needs instead of reading entire files.