Scaledown Claude Code Plugin

Optimize your Claude Code sessions with Scaledown — automatic context compression, conversation summarization, intent-aware tool routing, and named entity extraction.

What it does

Every time you submit a prompt, the plugin:

1. Classifies your intent and prepends a one-line hint (e.g. [Scaledown intent: file_read (87%)]) so Claude picks the right tool without guessing
2. Compresses large contexts automatically when you paste in a big codebase and ask a retrieval-style question — reducing token usage by 50–70% before the prompt reaches Claude

On top of that, Claude gains four new tools it can call on demand:

| Tool | What it does | |---|---| | sd_compress | Compress a large context block before a needle-in-a-haystack query | | sd_summarize | Abstractively summarize text — useful for compacting long conversations | | sd_classify | Classify text against custom labels (e.g. bug vs. feature vs. question) | | sd_extract | Extract named entities or structured data from any text |

Requirements

• Node.js 18 or later
• A Scaledown API key — get one free at scaledown.ai/dashboard
• One of: Claude Code, Cursor, or OpenAI Codex CLI

Installation

Supported clients: Claude Code · Cursor · OpenAI Codex CLI

The MCP tools (sd_compress, sd_summarize, sd_classify, sd_extract) work in all three clients. Automatic prompt hooks (UserPromptSubmit, PreCompact) are Claude Code-only — see the feature comparison below.

Claude Code

Option A: npm (recommended)

npm install -g @scaledown/claude-plugin
scaledown-claude setup

The setup wizard will:

1. Open your browser to get an API key
2. Ask you to paste the key
3. Save it to your shell config (~/.zshrc, ~/.bashrc, etc.)
4. Register the MCP server with Claude Code
5. Add the UserPromptSubmit hook to your project's .claude/settings.json

Restart Claude Code and you're done.

Option B: manual

1. Clone and build

git clone https://github.com/scaledown-team/scaledown-claude-plugin
cd scaledown-claude-plugin
npm install && npm run build

2. Set your API key

export SCALEDOWN_API_KEY="your-key-here"
# Add the above line to ~/.zshrc or ~/.bashrc to persist it

3. Register the MCP server

For personal use (stored in ~/.claude.json):

claude mcp add scaledown --transport stdio \
  -- node /path/to/scaledown-claude-plugin/dist/src/index.js

To share with your team (stored in .mcp.json, commit this file):

claude mcp add scaledown --transport stdio --scope project \
  -- npx -y @scaledown/claude-plugin

4. Add the hook

In .claude/settings.json at your project root (create if it doesn't exist):

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "scaledown-claude-hook"
          }
        ]
      }
    ]
  }
}

If you cloned the repo instead of installing globally, use the full path:

"command": "node /path/to/scaledown-claude-plugin/dist/hooks/user-prompt-submit.js"

Cursor

1. Install the package

npm install -g @scaledown/claude-plugin

2. Set your API key

export SCALEDOWN_API_KEY="your-key-here"
# Add the above line to ~/.zshrc or ~/.bashrc to persist it

3. Add the MCP server

Create .cursor/mcp.json in your project root (or ~/.cursor/mcp.json for global use):

{
  "mcpServers": {
    "scaledown": {
      "command": "npx",
      "args": ["-y", "@scaledown/claude-plugin"],
      "env": {
        "SCALEDOWN_API_KEY": "your-key-here"
      }
    }
  }
}

4. Restart Cursor. The four Scaledown tools will be available in Agent mode.

Cursor does not support hooks, so automatic prompt compression and intent classification will not fire. Use the tools on demand.

OpenAI Codex CLI

1. Install the package

npm install -g @scaledown/claude-plugin

2. Add the MCP server

codex mcp add scaledown --env SCALEDOWN_API_KEY=your-key-here -- npx -y @scaledown/claude-plugin

This writes to ~/.codex/config.toml. To verify:

[mcp_servers.scaledown]
command = "npx"
args = ["-y", "@scaledown/claude-plugin"]

[mcp_servers.scaledown.env]
SCALEDOWN_API_KEY = "your-key-here"

Codex CLI does not support UserPromptSubmit or PreCompact hooks. Use the tools on demand.

Feature comparison

| Feature | Claude Code | Cursor | Codex CLI | |---|---|---|---| | sd_compress tool | ✅ | ✅ | ✅ | | sd_summarize tool | ✅ | ✅ | ✅ | | sd_classify tool | ✅ | ✅ | ✅ | | sd_extract tool | ✅ | ✅ | ✅ | | Auto intent hints on every prompt | ✅ | ❌ | ❌ | | Auto compression (large prompts) | ✅ | ❌ | ❌ | | Auto summarization on compaction | ✅ | ❌ | ❌ |

Usage

Automatic (hook)

Nothing to do — the hook fires on every prompt. You'll see the intent hint in Claude's context, and large retrieval queries are silently compressed before they reach the model.

[Scaledown intent: search (82%)]
Find all places where we call the payments API

On-demand tools

Ask Claude to use any of the four tools directly:

Compress a large context

Use sd_compress to compress this before searching through it: [paste large codebase]

Summarize a long conversation

Use sd_summarize to condense this thread so we can keep working without hitting the context limit

Classify text

Use sd_classify to categorize these GitHub issues as bug, feature, or question

Extract structured data

Use sd_extract to pull out all function names, file paths, and error codes from this stack trace

Configuration

Changing your API key

Re-run setup to replace the key automatically:

scaledown-claude setup

Or edit your shell config directly:

# Open ~/.zshrc (or ~/.bashrc)
# Find and update:
export SCALEDOWN_API_KEY="sk-your-new-key"

# Reload
source ~/.zshrc

Environment variables

Set these environment variables to tune behavior:

| Variable | Default | Description | |---|---|---| | SCALEDOWN_API_KEY | — | Required. Your Scaledown API key | | SCALEDOWN_COMPRESS_THRESHOLD | 10000 | Token estimate above which auto-compression fires | | SCALEDOWN_COMPRESS_RATE | 0.3 | How aggressively to compress (0.3 = keep 30% of tokens) | | SCALEDOWN_NIAH_DISABLE | false | Set to true to compress all large prompts, not just retrieval-style ones |

Example — compress more aggressively, lower threshold:

export SCALEDOWN_COMPRESS_THRESHOLD=5000
export SCALEDOWN_COMPRESS_RATE=0.2

How compression works

The plugin uses a local heuristic to detect "needle-in-a-haystack" queries — prompts that are both large and retrieval-intent (containing keywords like find, search, where, what does ... do, etc.).

When detected, the full prompt is sent to Scaledown's /compress/raw/ endpoint, which rewrites it into a semantically equivalent but much shorter form. The compressed version replaces the original before Claude sees it.

Conversational messages that happen to be long (e.g. a big code block you're asking Claude to write from scratch) are left alone.

Development

git clone https://github.com/scaledown-team/scaledown-claude-plugin
cd scaledown-claude-plugin
npm install

npm test          # run unit tests
npm run build     # compile TypeScript

Test the hook manually:

npm run build
echo '{"prompt":"find the function that handles auth"}' \
  | SCALEDOWN_API_KEY=your-key node dist/hooks/user-prompt-submit.js

Test the MCP server starts:

SCALEDOWN_API_KEY=test echo '{}' | node dist/src/index.js

License

MIT