Tokless CLI
A CLI installer for eight token-saving tools: rules, hooks, MCP servers and a local proxy for Claude Code, Codex and Cursor.
- 370 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- A network script edits the configuration of several agents and installs command hooks and a local proxy for model traffic.
Install
- Works in
Manual install
curl -fsSL https://raw.githubusercontent.com/HoangP8/tokless/main/scripts/install.sh | bashThe first command in the Manual section for macOS and Linux. The README has a PowerShell variant for Windows. Read the script before running it.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- A network script edits the configuration of several agents and installs command hooks and a local proxy for model traffic.
This is third-party code. Review the repository files before installing.
What it does
Tokless installs and configures a set of eight projects: `principles`, `ponytail`, `caveman`, `rtk`, `headroom`, `codegraph`, `context-mode` and `projectmem`. Each is wired in the way that suits the agent: an instruction block, a hook on Bash commands, a plugin, an MCP server or a local HTTP proxy. Of the 29 tools exposed by three MCP servers, 12 reach the context; Tokless trims the rest with a filtering proxy. The `--agents` and `--tools` flags control scope, and `tokless doctor` reports the state of tools, agents and routes.
Who it is for. People who work with several agents (Claude Code, Codex, Cursor, Cline) and are tired of assembling token savings from scattered packages by hand.
Good fit when
- On a new machine or with a new agent you need a code index, output compression and project memory within minutes.
- Long logs and large files clog the context, and `context-mode` keeps them out.
- Several agents on one machine should receive the same rules and one set of MCP tools.
- Before changing configuration, the result is previewed with `--dry-run` without writing anything.
Not a fit when
- The agent configuration is hand-built and must stay untouched: the installer edits files of several agents.
- Running scripts from the network through `curl | bash` is forbidden.
- Only one specific tool from the set is needed: it is simpler to install on its own.
Example request
Show what Tokless would change for Claude Code and Codex by running `tokless --dry-run --agents claude,codex`, and describe the entries.Limitations
The installer changes agent configuration: it adds instruction blocks, hooks, plugins, MCP entries and a local `headroom` proxy at `127.0.0.1:8787` that compresses model requests and routes traffic. By default the proxy starts at login (`tokless proxy autostart off` disables it). The main command runs a script from the network through `curl | bash`, so read `scripts/install.sh` first. The repository bundles third-party projects, and their behavior and licenses are defined by their authors.
How to disable. Run `tokless disable --agents claude` to unwire chosen agents while keeping the tools, or `tokless uninstall` for full removal including binaries. Stop the proxy with `tokless proxy down`.
FAQ
Which tools are in the set?
Eight: `principles`, `ponytail`, `caveman`, `rtk`, `headroom`, `codegraph`, `context-mode` and `projectmem`. Limit the set with `--tools`, for example `--tools rtk,codegraph`.
Which agents are supported?
The README identifiers include `claude`, `codex`, `cursor`, `cline`, `copilot`, `opencode`, `antigravity`, `amp` and more. The `--agents` flag selects them.
How do I preview the changes?
Run `tokless --dry-run`: it shows every write and changes nothing. After installing, `tokless doctor` reports the state.
How do I update the tools?
`tokless update` updates the CLI and all tools, and `tokless self-update` updates only the Tokless binary.
What does the headroom proxy do?
It compresses large requests and routes model traffic with your own keys. Control it with `tokless proxy up`, `down` or `status`.
Discussion
No questions or reviews yet.
Related
CLI toolsToken Savior indexes code by symbol, keeps memory between sessions and rewrites noisy Bash commands so the agent spends fewer tokens
An installer for a set of hooks, commands and plugins that compress tool output and save tokens in Claude Code.
tokendoctor shows where Claude Code overpays in tokens, proposes settings fixes and rolls them back with one command
A local proxy with a terminal dashboard: shows how many tokens each Claude Code and Codex request sends and saves the prompts to files
Token Optimizer MCP
token-optimizer-mcp
Context compression and caching, token savings tracking per agent. Global npm install turns on Claude hooks and wraps commands; telemetry is off by default
A zero-dependency Python CLI: one catalog of MCP servers and skills for every agent, with only a compact index in context
Comparison with similar tools
Cards that share the main search query, and how they differ.
README in short
Tokless is described as wiring rather than running: after installation the tools act on their own. A table in the README shows how each of the eight tools connects to each agent.