claude-tui-boost

skill
Security Audit
Fail
Health Warn
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 7 GitHub stars
Code Fail
  • rm -rf — Recursive force deletion command in .github/workflows/plugin-pr-ci.yml
  • rm -rf — Recursive force deletion command in .github/workflows/publish-v0.1.0.yml
  • network request — Outbound network request in adapter_contract_audit.test.mjs
  • Hardcoded secret — Potential hardcoded credential in deployment-preflight.test.mjs
  • process.env — Environment variable access in first-connect.test.mjs
  • fs module — File system access in first-connect.test.mjs
  • Hardcoded secret — Potential hardcoded credential in first-connect.test.mjs
  • process.env — Environment variable access in host-compatibility.test.mjs
  • process.env — Environment variable access in runtime-environment.test.mjs
  • Hardcoded secret — Potential hardcoded credential in runtime-environment.test.mjs
  • spawnSync — Synchronous process spawning in scripts/check-deps.mjs
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

让 Claude 订阅额度更耐用。通过真实 Claude Code 交互式 TUI 接入 Magpie/OpenCode;Opus 5.5 单组对照测得约 1.8~2 倍额度利用效率。

README.md

English · 简体中文

Claude TUI Boost

Make your Claude subscription go further with an interactive TUI

Bring real interactive Claude Code to Magpie/OpenCode, with persistent sessions and safe MCP tool handoffs.
Persistent sessions, streamed responses and MCP tool handoffs—without modifying Magpie itself.

MIT License
Release
macOS / Linux

Why use it?

In one Claude Pro + Opus 5.5 comparison, the interactive route consumed 44–50% less 5-hour quota per API-equivalent workload than the tested headless claude -p route—roughly 1.8–2× effective quota efficiency under those conditions. See the measurements and methodology discussion →

The difference was observed for a specific workload, not across all models or accounts. It does not increase your subscription limits or guarantee savings; session continuity and prompt-cache behavior also affect results.

Headless workflow in the comparison With this plugin
Execution A fresh claude -p process per turn Genuine Claude Code TUI in tmux/PTY, with native session resume
Integration Direct print-mode calls A claude-tui provider behind a Magpie-compatible endpoint
Tool handoff Headless request lifecycle MCP tool calls and results bridged back to the calling Agent
Observed quota usage Reference baseline 44–50% less in the linked single-workload test

Built for agent workflows

  • Continue across turns: Reuse an interactive Claude session instead of treating every gateway turn as a fresh conversation.
  • Keep the tools working: Stream responses and hand off concurrent MCP tool calls/results to your Agent; handle cancellation and bounded recovery.
  • See your Claude usage: Reads 5-hour / 7-day usage through the selected profile's native Claude Code /usage command.
  • Install as a plugin: The bundled Go Runtime, MCP helper and tmux wrapper run alongside Magpie. No Magpie source patch is required.
Magpie-compatible Agent / SDK client
                │
                ▼
           Magpie gateway
                │
                ▼
      Claude TUI provider plugin
                │
                ▼
     Go Runtime + MCP / tmux bridge
                │
                ▼
       Interactive Claude Code

Quick start

Requirements: Magpie with GitHub plugin support; official Claude Code CLI 2.1.268+; macOS or Linux (arm64/x64). The login flow can prepare tmux; Bun comes from the Magpie host. Native Windows is not supported. Install on the machine running Magpie and Claude Code (including a VPS), not just on a remote client.

1. Install (current release: v0.2.14):

magpie plugin add github:Guuuu233/claude-tui-boost
magpie plugin login claude-tui

The main branch is kept at the latest published release, so Magpie shows an update in Plugins whenever a new one is out (or run magpie plugin update). Leave out a #tag: a tag never moves and Magpie would never offer an update. Updating restarts the plugin host, which restarts the managed runtime, so update between tasks. Installed earlier with #v0.2.x? Run the command above once to switch.

2. Connect Claude. In Magpie, open the Claude TUI provider and choose Auto install & sign in to Claude (shown in Chinese as “自动安装并登录 Claude”). Complete the official Claude browser login when prompted. A dedicated Claude profile is created; the advanced “Connect already signed-in Claude” option is also available.

3. Select the provider in your Magpie-routed client. Choose the configured Claude TUI model (typically sonnet, opus or haiku aliases), then test a normal response and a harmless tool round-trip. These aliases do not grant model access.

Important: Installing the plugin does not intercept a standalone claude -p command or automatically reconfigure the Claude Agent SDK. The client must explicitly route compatible requests through Magpie. Check First connection and VPS/SSH setup when needed.

Please read before installing: This is an unofficial experimental third-party plugin. Using a personal Claude subscription through other agents may violate applicable terms and can lead to account restrictions. The upstream proposal was declined. The plugin can launch Claude with reduced permission prompts; use a dedicated profile in a trusted environment. Technical and security details · Recovery limitations. For a formally authorized SDK integration, use Anthropic's official API.

Documentation

Guide What you'll find
First connection · Remote/VPS setup Login, browser authorization, tmux and remote installation
Compatibility · Magpie host compatibility Platforms, tested boundaries and host upgrades
Recovery Interrupted requests, 409/502, opt-in recovery and tool safety
Technical reference Advanced configuration, exact security effects, tests and builds
Release notes · Changelog What's new and what remains experimental

Current release: v0.2.14 accepts requests up to 32 MiB (long, image-heavy conversations) and answers larger ones with 413 rather than a retried 502. v0.2.13 replays a completed tool reply only for the very same request and answers an exhausted Claude allowance with 429 instead of 502. Since v0.2.12 each conversation is served from its own history: an interrupted or rewritten conversation is rebuilt rather than refused. Tool-replay boundaries in long, complex sessions remain experimental; see release notes.

Contribute

Questions and bug reports are welcome in Issues; improvements are welcome via Pull Requests. Please remove credentials and private transcripts from reports. If the project is useful, consider giving it a ⭐.

Reviews (0)

No results found