Packages

Self-hosted deep research tool that generates comprehensive research reports in about 2 minutes using configurable thinking and task LLM models. Combines multiple LLM providers (OpenAI, Anthropic, Google, Deepseek, Mistral, Ollama, etc.) with multiple search providers (Tavily, Searxng, Brave, Exa, Firecrawl). Includes MCP server support, knowledge graph generation, research history, and a multi-language web UI.

An MCP server that exposes tools/commands to let AI assistants call the Gemini CLI (via an MCP client such as Claude Code/Claude Desktop) for natural-language analysis and file/codebase understanding, including a sandbox mode for safer code execution/testing.

Terminal-based MCP client that connects local Ollama language models to MCP servers. Supports agent mode with iterative tool execution, multi-server connections, STDIO/SSE/Streamable HTTP transports, MCP prompts, human-in-the-loop tool approval, thinking mode, and conversation history export. This is a CLIENT, not an MCP server.

A simple Model Context Protocol (MCP) mock server for developing and testing MCP clients. It exposes configurable mock tools, prompt templates, and mock resources via StreamableHTTP or SSE using a YAML configuration file.

minio-server is the MinIO server package, providing an S3-compatible object storage service (typically used to run a self-hosted S3-compatible endpoint for buckets/objects).

Provides Penpot’s official Model Context Protocol (MCP) server and an accompanying Penpot plugin. The MCP server exposes tools that let an AI client retrieve, transform, and create design elements in a Penpot design file by routing requests through a WebSocket-connected Penpot plugin.

Provides an MCP server that lets Claude Desktop ingest customer feedback from sources (Slack, Google Drive, Zoom, JIRA, CSV/manual), run AI analysis (clustering/insights/embeddings), score feedback using VOC dimensions, and generate/export evidence-backed PRDs with JIRA sync support.

An MCP (Model Context Protocol) server that exposes STAC (SpatioTemporal Asset Catalog) capabilities to agent tools, enabling LLM agents to search and retrieve STAC catalog/resource information via MCP.

Unity editor/MCP-oriented plugin that enables inspecting and modifying Unity ParticleSystem components via natural-language commands. It exposes MCP tools to get ParticleSystem data and to modify ParticleSystem modules/properties (24 supported modules).

Unity editor plugin that integrates MCP-compatible AI assistants with the Unity editor by running a WebSocket client inside Unity and dispatching tool calls to Unity-side command handlers (with Undo/Redo and reconnection handling).

Provides an npm-installed Node/TypeScript CLI + long-running daemon that manages MCP servers and exposes “progressive disclosure” layers (metadata/status → tool list → per-tool schema) to help agents call MCP tools with less context/token usage. Uses a Socket (newline-delimited JSON) between CLI and daemon, which then connects to MCP servers via MCP transports (e.g., stdio).

AllVoiceLab-MCP is an MCP server that exposes tools for text-to-speech, voice conversion, voice cloning, video translation/dubbing, subtitles extraction, and subtitle removal by calling AllVoiceLab’s underlying TTS/VC/video translation APIs.

google-slides-mcp is an MCP (Model Context Protocol) server that provides tool functions to interact with the Google Slides API, enabling an agent to create presentations, fetch presentation/slide data, perform batch updates to modify slides, and summarize slide text content.

An MCP (Model Context Protocol) server that exposes Hacker News data as callable tools. It provides tools to fetch news lists (top/new/best), fetch story details with comments, fetch user profiles/recent activity, and search Hacker News via the Algolia API; it includes input validation with zod and HTTP calls via axios.

IronCurtain is a TypeScript runtime/CLI for autonomous AI agents that enforces a human-readable “constitution” (policy) compiled into deterministic rules. It mediates all agent tool calls via MCP servers (e.g., filesystem/git/github/workspace) and a policy engine that allows/denies/escalates actions for user approval, with agent code isolated in a V8 sandbox (builtin mode) or an external agent constrained by Docker + network/MCP mediation (docker mode).

A lightweight Flask API bridge that connects Claude Desktop (or any MCP-compatible client) to a Kali Linux machine, enabling AI-assisted command execution for authorized penetration testing and CTF challenges. The server exposes a thin HTTP API on port 5000 that accepts arbitrary shell commands and returns their output, effectively giving an AI assistant a live Kali terminal. Tools like Nmap, Metasploit, sqlmap, Gobuster, enum4linux, and any other Kali tool are accessible by name. The architecture is intentionally minimal — a single Flask server with no auth, no sandboxing, and no command filtering — making it fast to set up for isolated testing environments but completely unsuitable for production or shared infrastructure.

A Spring Boot–based MCP (Model Context Protocol) server that exposes a tool to query local computer/system configuration (OS info, hardware details, current user info, and Java runtime version), intended to run via stdio in an MCP client.

Provides a Go-based CLI tool and an MCP server for writing notes to Flomo, with support for Markdown content and comma-separated tags. Configuration is done via an environment variable pointing at a Flomo API/endpoint URL.

Provides a Python MCP (Model Context Protocol) server with coding-plan specific tools, including web search and image understanding, intended to be used by MCP-compatible coding assistants via stdio (local) or SSE (network/cloud) transports.

Provides an MCP server with multiple tools for searching and analyzing US patent data using the free USPTO PatentsView API (e.g., search, patent details, citations, technology trends, CPC/IPC classification search, families, claim extraction, top holders, portfolio comparison, and landscape/strategic analysis).