Coding agent plugin · v0.5.40

One install. Every coding agent. Fewer tokens.

Run supercompress setup once. It links your account, detects coding agents, and installs a clean MCP + hooks plugin across your coding agents: Cursor, Claude Code, Codex, Goose, Zed, OpenCode, fx, Hermes, OpenClaw, and more. Keep your normal login. Check health anytime with supercompress doctor.

Install in 30 seconds
npm install -g supercompress-proxy
supercompress setup
supercompress doctor

setup links your account and wires MCP + hooks on every detected host. Re-run supercompress plugin anytime. doctor prints a clean per-agent health matrix.

query-awarekeeps the context that answers your question
hostedquery-aware compression API
extractiveoriginal lines kept verbatim

First 5 minutes checklist

Get SuperCompress running with your coding agents in 4 quick copy-paste steps:

1
Install & link account

Install the CLI globally and run one-time setup to connect your account and auto-configure detected agents:

npm install -g supercompress-proxy
supercompress setup
2
Verify agent detection

Confirm which coding agents on your machine were detected and hooked:

supercompress agents
3
Test MCP connectivity

Verify that the SuperCompress MCP server is responsive locally:

supercompress mcp-check
4
Restart agent & compress

Restart Cursor, Claude Code, Codex, Roo Code, or Cline so the MCP server initializes. Your agent will now automatically call compress_context on bulky context dumps!

How it works

Two layers, installed together by setup:

  1. Hooks (Cursor / Claude Code / Codex), every user message + large tool dumps auto-compress.
  2. MCP on every detected agent, compress_context for big dumps (subscription / login safe).
# After `supercompress setup` (subscription / login safe)
Coding agent ──→ hooks / compress_context ──→ SuperCompress API
Auth Laptop setup uses browser device-link (Google). Hosted MCP at /api/mcp accepts an sc_ API key header. OAuth 2.1 + PKCE is how a keyless Google sign-in path is designed; it is not the recommended install today. Context is compressed on the SuperCompress API; provider keys never leave your agent.

Install

npm install -g supercompress-proxy
supercompress --version
# → 0.5.40

Or one-shot without a global install:

npm install supercompress-proxy
npm exec supercompress setup

Detect + plugin

setup links your account and installs the MCP plugin. Prefer plugin anytime you add a new agent:

supercompress setup
supercompress plugin
$ supercompress plugin
✓ Detected Cursor, Claude Code, Codex, FreeBuff, OpenCode, …
✓ MCP plugin installed for: Cursor, FreeBuff, OpenCode, Codex MCP, …
✓ Cursor rule written
→ Works with login, no provider API-key mode
→ Restart your coding agent so MCP reloads
1
Install

Add supercompress-proxy with npm.

2
Link

supercompress setup opens the dashboard and connects your account.

3
Restart

Reload Cursor / Claude / Codex / FreeBuff / OpenCode so MCP servers load.

4
Compress

Ask the agent to compress large context, tool: compress_context.

Supported agents

Works across most coding-agent harnesses. The popular ones get a first-class auto MCP plugin from setup / plugin, native config shape per host, not a clunky one-size paste:

Cursor

~/.cursor/mcp.json + hooks + always-on rule.

MCP

Claude Code

~/.claude.json + hooks + CLAUDE.md.

MCP

Codex

~/.codex/config.toml MCP + hooks.

MCP

Goose

YAML extensions.supercompress stdio MCP.

MCP

OpenCode

mcp.supercompress in opencode.json(c).

MCP

Zed

context_servers.supercompress (native).

MCP

Grok / fx / Hermes / OpenClaw

Host-native MCP + instructions / skills.

MCP

Windsurf · Continue · Cline · Roo · Gemini · FreeBuff · Claude Desktop · …

Auto-written into each host’s MCP schema.

MCP

Also catalogued, supercompress agents connect or OpenAI/Anthropic base URL

AiderOpenHandsAmazon QJetBrains AITraeAntigravityAugmentWarpDevinCopilot CLIPiCrush+ more

Run supercompress agents for the full list on your machine. Health: supercompress doctor.

MCP tools

ToolWhat it does
compress_contextCompress bulky coding context for a query. Returns compressed text + savings.
connect_accountOpen the dashboard to link this install if you skipped setup.
usage_summaryPer-agent token savings for the connected SuperCompress account.

Dashboard tracking

After agents start compressing, your dashboard shows per-agent requests, tokens in/out/saved, and last-used time. Only usage metadata is sent, never prompt contents.

Optional API proxy

If you want the older base-URL rewrite path (provider API-key mode), opt in explicitly:

supercompress setup --proxy
supercompress start

That wires http://localhost:8080/v1 for compatible OpenAI/Anthropic endpoints. Most subscription/login agents should stay on the default MCP plugin path instead.

When to use --proxy Only when your agent already uses a provider API key and exposes a configurable OpenAI/Anthropic base URL. ChatGPT-login Codex and similar hosted backends are not intercepted by the local proxy, use MCP.

Uninstall

supercompress uninstall
npm uninstall -g supercompress-proxy

Uninstall removes MCP registrations, restores backed-up configs, and cleans ~/.supercompress.