Odel
toolpiper

toolpiper

Local
@modelpiper2JavaScriptMITUpdated Yesterday

On-device MCP tools for macOS: local AI, speech, vision, browser automation, system control.

ToolPiper MCP Server for macOS

400+ Model Context Protocol tools, all running on-device. ToolPiper is a native macOS app that turns your Mac into a local AI platform: bundled inference (llama.cpp on Metal, Apple Intelligence, speech on the Neural Engine), an OpenAI-compatible API at localhost:9998, and a built-in MCP server that exposes the whole machine to any MCP client: Claude Code, Cursor, Codex, or your own scripts.

No cloud. No Docker. No Python. No API keys. Nothing leaves your Mac, and you can verify that yourself: lsof -i -P | grep ToolPiper while it runs.

This repository is the official home for ToolPiper's MCP server manifest (server.json), the Claude Code plugin (plugins/toolpiper), and the setup guides. The ToolPiper app itself is not open source, but its privacy claims do not ask for trust: every byte of network traffic is observable on your own machine. Open-source siblings live in this org: PiperSR (ANE super-resolution model), FluidAudio, mlx-audio-swift.

Quick start

Install ToolPiper, launch it once, then connect any MCP client over Streamable HTTP. Loopback connections need no auth: no token, no OAuth step.

Download the disk image from modelpiper.com/download, or install with Homebrew:

brew tap modelpiper/tap
brew trust modelpiper/tap
brew install --cask toolpiper

The trust line is required once: Homebrew 6 refuses casks from taps outside its own until you trust them.

Claude Code, as a plugin (one server entry, nothing to configure by hand):

claude plugin marketplace add ModelPiper/toolpiper
claude plugin install toolpiper@modelpiper

Claude Code, as a manual server:

claude mcp add --transport http toolpiper http://127.0.0.1:9998/mcp

Any other MCP client, in its server configuration:

{
  "mcpServers": {
    "toolpiper": { "type": "http", "url": "http://127.0.0.1:9998/mcp" }
  }
}

Claude Desktop, as a one-click extension: the packaged .mcpb ships with the next release. Source, build script and privacy statement are here now: desktop-extension/. It is a small Node.js forwarder that relays Claude Desktop's stdio to the same loopback server, and it serves the directory profile: the catalog minus every tool that uses an AI model to generate images, video or audio.

Installing with an AI agent? Point it at llms-install.md.

What's in the catalog

400+ tools, 5 resources, and 5 workflow prompts. Names follow <category>_<action> (vision_ocr, system_window_snap, browser_assert). The live catalog, with per-tool documentation, is at modelpiper.com/mcp-tools.

CategoryExamplesWhat it covers
Core AIchat, audio_transcribe, audio_speak, text_embedLocal LLM inference, speech-to-text and text-to-speech on the Neural Engine, embeddings
Apple frameworks + RAGvision_ocr, image_analyze, text_analyze, rag_queryApple Vision OCR, image analysis, natural-language analysis, hybrid HNSW + BM25 document search
Browser automationbrowser_snapshot, browser_action, browser_assert, browser_interceptChrome via CDP: AX-tree snapshots, self-healing selectors, network mocking, Web Vitals, coverage
Testingtest_run, test_save, test_exportSaved browser test sessions, fast replay, Playwright/Cypress export
System actionssystem_*Windows, Spaces, clipboard (with OCR'd history), audio, displays, Finder, Calendar, Reminders, Contacts, notifications, processes, network, Bluetooth, Shortcuts, and more
Video creatorvideo_record, video_render, video_edit_timelineScreen-recorded screenplays, non-destructive timeline editing, OTIO/FCP XML/EDL export
Outreachhn_search, reddit_search, github_compare, gsc_analyticsHN, Reddit, GitHub, Google Search Console
Model managementmodel_search, model_download, model_loadSearch HuggingFace, download GGUF/CoreML, load/unload
Routing + integrationsendpoint_set, endpoint_recommend, claude_code_installSwitch inference endpoints per session, Claude Code helper install

Resources: toolpiper://status, toolpiper://models, toolpiper://backends, toolpiper://tests, toolpiper://categories Prompts: setup-model, audit-page, create-test, voice-over, research-topic

Category and profile filtering

Don't want the whole catalog in context? Pass category or profile as a query parameter on the /mcp URL. Both the tool list and the server instructions trim to match:

# Just LLM inference (core category)
claude mcp add --transport http toolpiper "http://127.0.0.1:9998/mcp?category=core"

# Web dev workflow (automation profile: core + browser + testing)
claude mcp add --transport http toolpiper "http://127.0.0.1:9998/mcp?profile=automation"

Transport

TransportEndpointAuth
Streamable HTTPPOST http://127.0.0.1:9998/mcp (plus GET /mcp, the SSE channel for tools/list_changed)None for loopback peers, enforced at the kernel socket, so Host/Origin spoofing cannot widen it. LAN and remote peers present Authorization: Bearer.

Streamable HTTP is the only transport. ToolPiper has no stdio server, and older guides that described one are obsolete.

Privacy

ToolPiper never calls out. No telemetry, no analytics, no crash uploads, no account requirement. The only network traffic is the traffic you ask for (model downloads from HuggingFace, tools that explicitly target the web). This is a property you can check, not a policy you have to trust:

lsof -i -P | grep ToolPiper

Requirements and pricing

  • macOS 26 or later on Apple Silicon
  • Free: the full local model runner and every speech feature (transcription, text to speech, dictation). No caps, no account.
  • Paid plans add more of the catalog. The current ladder is at modelpiper.com/pricing.

FAQ

Is this the same as "ToolPipe"? No. ToolPiper (this page, modelpiper.com) is the macOS MCP server by ModelPiper. Unrelated projects with similar names exist in some directories.

What does the MIT license in this repo cover? The contents of this repository: the server manifest, the Claude Code plugin, and the setup guides. The ToolPiper app itself is proprietary.

Does it work without an internet connection? Yes. Inference, speech, vision, RAG, and system tools are all on-device. You only need the network to download models.

Can other machines connect? Loopback is zero-config. LAN and remote access go through bearer-token auth.

I already added ToolPiper with claude mcp add. Do I need the plugin? No. Claude Code keeps your manual toolpiper server and skips the plugin's copy while the manual one exists, so nothing doubles up. The plugin is for a fresh setup, or if you remove the manual entry.