cacheout-mcp
MCP server for macOS disk cache management. Enables AI agents to free disk space on demand by scanning and clearing developer caches.
README
cacheout-mcp
MCP server for macOS disk cache management — lets AI agents free disk space on demand.
Why
When you run AI agents (OpenClaw, Claude Code, etc.) on a Mac Mini or any macOS machine, disk pressure can silently degrade performance. Swap thrashing, failed builds, Docker OOM — all from running out of space that's locked up in developer caches.
cacheout-mcp gives your agent a set of tools to detect and fix disk pressure in real-time:
Agent detects 2 GB free → calls cacheout_smart_clean(target_gb=15) → 15 GB freed in 3 seconds
Three Execution Modes
| Mode | When | How it works |
|---|---|---|
| Socket | Cacheout daemon running with Unix socket | Connects to daemon for real-time data, trend analysis, health scores |
| App | Cacheout.app installed, no running daemon | Delegates to Cacheout CLI binary (--cli flag) |
| Standalone | No Cacheout.app (headless servers) | Cleans caches directly via Python, reads sysctl for memory stats |
Mode is auto-detected at startup (socket → app → standalone). Override with CACHEOUT_MODE=standalone or CACHEOUT_MODE=app.
Install
# From PyPI (when published)
pip install cacheout-mcp
# From source
cd cacheout-mcp
pip install -e .
MCP Configuration
Claude Code / Claude Desktop
Add to ~/.claude/claude_desktop_config.json or your project's .mcp.json:
{
"mcpServers": {
"cacheout": {
"command": "cacheout-mcp",
"env": {}
}
}
}
OpenClaw / Custom Agents
{
"mcpServers": {
"cacheout": {
"command": "python",
"args": ["-m", "cacheout_mcp.server"],
"env": {
"CACHEOUT_MODE": "standalone"
}
}
}
}
With uv (no install needed)
{
"mcpServers": {
"cacheout": {
"command": "uvx",
"args": ["cacheout-mcp"]
}
}
}
Tools
cacheout_get_disk_usage
Check current disk space. No parameters.
→ {"total": "500.1 GB", "free": "23.4 GB", "used_percent": 95.3}
cacheout_scan_caches
Scan all cache directories and report sizes. Optional filters:
categories: List of slugs to scan (omit for all)min_size_mb: Only show categories above this size
→ {"total_cleanable": "45.2 GB", "categories": [{"slug": "xcode_derived_data", "size_human": "15.0 GB", ...}]}
cacheout_clear_cache
Clear specific categories by slug. Requires explicit category list.
categories: Required list of slugsdry_run: Preview without deleting
← {"categories": ["xcode_derived_data", "homebrew_cache"], "dry_run": false}
→ {"total_freed": "18.2 GB", "results": [...]}
cacheout_smart_clean
The primary tool for agents. Specify how much space you need; it clears safest caches first.
target_gb: Required — how many GB to freedry_run: Preview modeinclude_caution: Include Docker and other high-risk categoriesfree_memory: Also run memory purge after disk cleanup (addsmemory_freed,purge_resultto response)
← {"target_gb": 10.0}
→ {"target_met": true, "total_freed_human": "12.3 GB", "disk_after": {"free_gb": 17.5}}
cacheout_status
Server status, mode, and available categories.
cacheout_get_memory_stats
Check RAM, swap, memory pressure, and memory tier. No parameters.
→ {"total_physical_mb": 16384.0, "memory_tier": "comfortable", "estimated_available_mb": 6096.0, ...}
cacheout_get_process_memory
List top memory-consuming processes. Optional top_n and sort_by parameters.
Returns envelope: {mode, capabilities, data: {processes, count, sort_by_applied}, partial}.
cacheout_get_compressor_health
Check macOS memory compressor ratio, compression/decompression rates, and thrashing detection. No parameters.
Returns envelope: {mode, capabilities, data: {compressor_ratio, thrashing, ...}, partial}.
cacheout_memory_intervention
Run memory interventions. Required parameters:
intervention_name: Canonical name (e.g.,"purge")confirm:falsefor dry-run preview,trueto execute
Returns envelope: {mode, capabilities, data: {dry_run, intervention, ...}, partial}.
cacheout_system_health
Combined disk + memory + alert health check with a 0-100 score. No parameters. In socket mode, fetches from daemon. In standalone, computes locally.
cacheout_check_alerts
Read watchdog alerts (near-zero cost file read). Optional acknowledge parameter.
cacheout_get_recommendations
Get predictive memory/disk recommendations. Socket mode includes trend-based types
(exhaustion_imminent, compressor_degrading). Standalone returns snapshot types only.
cacheout_configure_autopilot
Validate and apply autopilot/watchdog configuration. Required config parameter (dict with
version, enabled, optional rules/webhook/telegram). This is a write/validate/apply tool.
Cache Categories (23 total)
| Slug | Risk | What it cleans |
|---|---|---|
xcode_derived_data |
Safe | Build artifacts, indexes |
uv_cache |
Safe | Fast Python installer cache |
homebrew_cache |
Safe | Downloaded bottles/tarballs |
npm_cache |
Safe | npm package cache |
yarn_cache |
Safe | Yarn package cache |
pnpm_store |
Safe | pnpm content-addressable store |
bun_cache |
Safe | Bun package manager cache |
typescript_cache |
Safe | TypeScript compiler + Next.js SWC cache |
playwright_browsers |
Safe | Playwright browser binaries |
cocoapods_cache |
Safe | CocoaPods specs and pods |
node_gyp_cache |
Safe | Native Node.js addon headers |
prisma_engines |
Safe | Prisma ORM query engine binaries |
swift_pm_cache |
Safe | Swift Package Manager cache |
gradle_cache |
Safe | Gradle build cache |
pip_cache |
Safe | Python pip cache |
chatgpt_desktop_cache |
Safe | ChatGPT desktop app cache |
vscode_cache |
Safe | VS Code updates and extensions |
electron_cache |
Safe | Shared Electron framework cache |
browser_caches |
Review | Brave/Chrome cached web content |
xcode_device_support |
Review | iOS device debug symbols |
simulator_devices |
Review | iOS/watchOS simulator data (uses xcrun) |
torch_hub |
Review | PyTorch models (slow to re-download) |
docker_disk |
Caution | Docker virtual disk (all images/containers) |
Smart Clean Priority
When smart_clean is called, categories are cleaned in this order:
- Safe categories, sorted by clean_priority (build artifacts first)
- Review categories (browser caches, device support)
- Caution categories (Docker) — only if
include_caution=true
Stops as soon as target_gb is freed.
Environment Variables
| Variable | Default | Description |
|---|---|---|
CACHEOUT_MODE |
auto-detect | Force standalone or app mode |
CACHEOUT_BIN |
auto-detect | Path to Cacheout binary |
Adding the CLI to Cacheout.app
If you maintain the Cacheout Swift app, add the CLIHandler.swift file to your Sources
and update CacheoutApp.swift to check CLIHandler.shouldHandleCLI() on init.
This enables Cacheout --cli scan, Cacheout --cli clean, etc. for app-mode integration.
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.