Open source · MIT licensed · Local-first

Tooling for local-first AI development.

MCP servers, command-line tools, training and dataset pipelines, and game-production tooling from MCP Tool Shop. Built to run on your own hardware.

86 classified repositories
58 installable packages
50+ releases in the last 90 days
MIT every repository

Featured tools

Installable today. Each has tagged releases, CI, and a support policy.

All tools

Tool Compass

stable

Find MCP tools by describing what you need. Hybrid semantic + lexical search.

npx @mcptoolshop/tool-compass
MCP server Python

Drive ComfyUI from Python: image, video, 3D and audio graphs with no node canvas.

pip install comfy-headless
CLI Python

Headless LLM fine-tuning in three lines, with VRAM-aware defaults and GGUF export.

npx @mcptoolshop/backpropagate
Library Python

Text-to-speech MCP server: 48 voices, 9 languages, SSML-lite, multi-speaker dialogue.

pip install voice-soundboard
MCP server Python

Delegate bulk analysis from Claude Code to a local Ollama model. 42 job-shaped tools.

npx ollama-intern-mcp
MCP server TypeScript

Shipcheck

stable

Release quality gate: Ship Gate checklist, error contract, templates, and audit.

npx @mcptoolshop/shipcheck
CLI JavaScript

Recent releases

Tagged on GitHub and published to npm or PyPI from CI.

All releases

Start here

Five paths. Each names three or four repositories. You do not need the rest.

All paths

What we build

Seven areas. The names below are starting points; every public repository is classified.

Ecosystem spec

MCP servers

10

Servers and gateways that speak the Model Context Protocol, plus the tooling that tests them.

  • Tool Compass Find MCP tools by describing what you need. Hybrid semantic + lexical search.
  • Ollama Intern MCP Delegate bulk analysis from Claude Code to a local Ollama model. 42 job-shaped tools.
  • MCP Voice Soundboard Text-to-speech MCP server: 48 voices, 9 languages, SSML-lite, multi-speaker dialogue.
  • Polyglot MCP Local GPU translation MCP server. TranslateGemma via Ollama, 55 languages.
  • Plain Sight Point it at an image, an AI says what it sees. Florence-2 describer — MCP server + CLI (local, MIT)
  • AI Eyes MCP Grounded visual evaluator MCP server — honest image judgment via SigLIP2 on a pinned model revision. Measures one image-text pair, names the weights that produced the score, and abstains rather than guessing.

Training and datasets

9

The local-model lifecycle: fine-tuning, datasets, training-step proof, and hardware capacity planning.

  • Backpropagate Headless LLM fine-tuning in three lines, with VRAM-aware defaults and GGUF export.
  • Style Dataset Lab Canon-aligned visual dataset production: snapshots, splits, training packs, evals.
  • Repo Dataset Turn code repos and curated visual repos into contamination-checked training datasets…
  • Backprop Trace Deterministic 26-rule verifier for neural-network training steps.
  • RunForge VS Code Push-button ML training with presets, indexed runs, and artifact inspection — all…

Image and media pipelines

15

Programmatic generation and verification of images, 3D, video and audio.

  • Comfy Headless Drive ComfyUI from Python: image, video, 3D and audio graphs with no node canvas.
  • Comfy Preflight A gate that runs on a ComfyUI workflow graph in the seconds before submission and halts a submission that would spend credits producing a known-wrong result.
  • Sprite Foundry Headless sprite generation pipeline: ComfyUI, SQLite, and a Godot finish lab.
  • Armature Image-to-video with a GLB instead of an image.
  • Audiobooker EPUB, TXT, PDF, or DOCX to multi-voice narrated audiobooks with mastering.
  • CodeComfy VS Code VS Code extension for code analysis and coverage display.

Agent infrastructure

15

What a coding agent runs on: orchestration, context routing, knowledge substrates, verification, health.

  • Role OS Multi-Claude orchestration: 61 roles, team packs, missions, and structured evidence.
  • Loadout OS Knowledge OS for AI coding agents — one CLI that routes the right context (memories +...
  • Prism Verify Runtime LLM verifier — family-different, reasoning-stripped, multi-lens adjudication with signed Ed25519 receipts (CLI · MCP · HTTP).
  • Claude Guardian Flight computer for Claude Code — log rotation, watchdog, crash bundles, and MCP…
  • Research OS Local-first research control plane for gated source packs, claim truth, contradiction…

Developer tooling

13

Release gates, registries, site scaffolding and the machinery of shipping.

  • Shipcheck Release quality gate: Ship Gate checklist, error contract, templates, and audit.
  • Site Theme Shared Astro theme for MCP Tool Shop org repo landing pages.
  • Repo Knowledge Repo knowledge system — relational catalog with full-text search, audit evidence, and…
  • Registry Stats Multi-registry download stats for npm, PyPI, NuGet, VS Code Marketplace, and Docker Hub.
  • npm Launcher Generic GitHub-release binary launcher with cache + SHA256 verification.
  • Forkctl Adoption control plane for GitHub repos.

Games and game tooling

18

Shipped games and the authoring tools that produce them.

  • AI RPG Engine Simulation-native toolkit for building, analyzing, and balancing deterministic RPG worlds.
  • World Forge 2D world authoring studio for AI RPG Engine.
  • Motif Adaptive soundtrack studio for composing, arranging, scoring, and exporting interactive...
  • Saints Mile A frontier JRPG for the adults who loved those games first
  • Roll RPG dice engine — expression parser, probability analyzer, loot tables, beautiful…
  • Star Freight Star Freight — Space Merchant RPG

Ledger and verification

6

Attestation, receipts and on-ledger proof for releases and financial truth.

  • Attestia Financial truth infrastructure for decentralized systems.
  • XRPL Lab XRPL training workbook. Learn by doing, prove by artifact.
  • XRPL Camp Learn the XRP Ledger in one sitting — real transactions, portable proof, 10 minutes.
  • XRPL Creator Capsule Creator-owned release system on XRPL — desktop app + CLI for publishing, minting, access gating, and recovery. Testnet preview.
  • Repomesh Trust infrastructure for repo networks — verify releases, check anchors, and onboard…

Testing, from dogfood-lab

An open workshop for testing in the AI era: methods, operating systems, and showcases for how AI-assisted software should be verified.

All dogfood-lab repositories

testing-os

v1.12.0

Operating system for testing in the AI era: protocols, evidence stores, and learning loops for AI-assisted software.

npx @dogfood-lab/dogfood-swarm
TypeScript Handbook

study-swarm

v2.0.0

Ground design decisions in cited research, then verify every citation with a different model family before it becomes canon.

npx @dogfood-lab/study-swarm
TypeScript Handbook

How we ship

Local-first

Core functionality never needs an API key or a hosted service. Where cloud routing exists it is opt-in and falls back to local.

MCP-native

Servers work with Claude Code, Claude Desktop, Cursor, VS Code, and any other Model Context Protocol client.

Gated releases

Every tagged release passes the shipcheck gate: security policy, structured errors, current docs, clean packaging.

Classified

Every public repository has an area, a lane, and a verified install command in the ecosystem specification.