mcp-agent is a Python framework for building, deploying, and serving AI agents that interact with the Model Context Protocol (MCP). In production, it provides a library and CLI toolkit that allows developers to compose agents using configurable LLM providers (OpenAI, Anthropic, Azure, Bedrock, Ollama, Google), connect to any MCP server (stdio, SSE, Websocket, Streamable HTTP), and orchestrate multi-step workflows with patterns like parallel fan-out/fan-in, router, orchestrator-worker, evaluator-optimizer, and swarm. The framework is architecture-agnostic, supporting both asyncio in-process execution and durable execution via Temporal for long-running, fault-tolerant agent loops. The system ships with a feature-rich CLI (mcp-agent) for local scaffolding, configuration, and cloud deployment to MCP Agent Cloud, including secret management, log tailing, and OAuth integration. Key entry points include the main application bootstrap (MCPApp), the Agent class (managing MCP server connections and LLM), workflow implementations in src/mcp_agent/workflows/, and the server module that exposes agents as MCP servers via FastMCP. The codebase is well-structured with clean abstractioons (ContextDependent, AugmentedLLM base), extensive testing (over 100 test files), and comprehensive observability through OpenTelemetry tracing, token counting, and structured logging. Use cases range from simple chat agents to complex research assistants, real estate analyts, and customer service desks with swarm routing. The framework is actively maintained with frequent commits, a large community, and is built to production standards.
MCP-UI is a monorepo SDK implementing the MCP Apps standard for rendering interactive UIs over the Model Context Protocol. It provides server-side tools (in TypeScript, Python, and Ruby) to create UI resources, and a client-side React SDK to render them in sandboxed iframes. The architecture centers on a secure double-iframe proxy pattern that isolates third-party HTML and external URLs, with JSON-RPC communication via AppBridge. The codebase is well-structured with comprehensive tests, semantic release automation, and cloudflare workers example, demonstrating strong engineering practices across multiple languages.
This repository is a production-grade Model Context Protocol (MCP) server that bridges AI assistants to Figma. It exposes 100+ MCP tools for reading and manipulating Figma files, variables, components, styles, FigJam boards, Slides, version history, tokens, and accessibility audits. The architecture is dual-mode: a local Node.js server (`src/local.ts`) communicates over stdio to AI clients and via WebSocket to a bundled Figma Desktop Bridge plugin, while a Cloudflare Workers variant (`src/index.ts`) supports cloud sessions with OAuth and a Durable Object WebSocket relay. The Desktop Bridge is the primary data channel, with Figma REST API fallbacks for scenarios where the plugin is unavailable or limited to certain plans. Beyond basic Figma CRUD, the codebase contains substantial engineering investment in design-token synchronization: canonical DTCG-aligned token documents round-trip through parsers and formatters for CSS variables, Tailwind v3/v4, SCSS, TypeScript modules, Style Dictionary, and Tokens Studio. A separate extraction engine (`src/core/extraction/`) mines production codebases for styling conventions and scaffolds design-system packages, while a scoring engine produces weighted design-system health dashboards. These components give the repo a hybrid quality: it is fundamentally an application, but its token pipeline and codebase-analysis subsystems are data-engineering-like assets that could be reused independently. The codebase demonstrates strong engineering practices: strict Zod schemas, consistent tool registration factories, connector abstraction, extensive Jest test coverage, atomic file writes, adaptive response compression, and careful caching/invalidation. However, it also carries meaningful debt: some token parsers are unimplemented stubs, a few files are monolithic (`figma-tools.ts`, `src/local.ts`), local/cloud connector implementations must be kept in sync manually, and some documentation references outdated internal paths. Overall, this is a mature, feature-rich MCP server with several genuinely reusable components for anyone building design-tool pipelines, token bridges, or local process-lifecycle infrastructure.
This is an MCP (Model Context Protocol) server implementation that provides LLMs with MySQL database access. The server exposes a single `mysql_query` tool for executing SQL queries and resources for listing tables and column metadata. It supports both stdio and HTTP transport modes, the latter with bearer token authentication. Key features include multi-database mode (querying across schemas without reconnection), schema-specific write permissions (INSERT/UPDATE/DELETE/DDL per database), SSL/TLS with mTLS, Unix socket connections, and a sophisticated PII redaction system that combines column-name heuristics with regex value scanning and Luhn-verified credit card detection. The architecture is monolithic: a single Express server (or stdio process) manages a mysql2 connection pool and routes all MCP protocol requests through zod-validated schemas. The codebase is well-tested with unit, integration, and end-to-end test suites, and demonstrates careful engineering in areas like query parsing (using node-sql-parser), permission enforcement, and secure defaults (write operations disabled by default).
Operately is an open-source company operating system for coordinating goals, projects, and teams. It provides a full collaboration suite including spaces (groups), goals with OKRs, project management with milestones and tasks, check-in cadence, retrospectives, discussions, document and file management (Resource Hubs), KPI tracking, and people management. The system is built as a monolith with a Phoenix (Elixir) backend serving a custom TurboConnect API (mutation/query protocol) and a React/TypeScript single-page application frontend with a reusable component library (turboui). It also offers a CLI tool, Model Context Protocol (MCP) integration for AI agents, and extensive background job processing via Oban for notifications, email delivery, search indexing, and billing operations. The architecture follows a clean domain-driven design with operations (Ecto.Multi transactions), activity logging (event sourcing with embedded schemas), an access control system with hierarchical groups and bindings, and a rich content editor with link enrichment and blob storage. The codebase is well-tested with both backend (ExUnit, Wallaby) and frontend (Jest, React Testing Library) tests, and includes a demo data builder for development. The enterprise edition (ee/) extends the core with SaaS admin capabilities, billing management via Polar, site messages, and feature flags. The system has evolved significantly with over 115 data migrations, indicating active development. While the core architecture is solid and many patterns are reusable, there are some incomplete features (e.g., stubbed notification handlers, missing email templates) and occasional code duplication between internal API and MCP tool layers. Overall, it demonstrates strong engineering craftsmanship with practical trade-offs for an open-source SaaS application.
An Obsidian plugin (Gemini Scribe) that integrates multiple AI providers (Gemini, Anthropic, OpenAI, Ollama) for chat, agent-based automation, and knowledge management within the Obsidian vault. The system provides a modular architecture with an agent loop that executes tools (vault operations, web search, RAG semantic search, image generation, deep research) with confirmation flows and diff previews. It features a declarative settings UI with per-feature provider routing, a skill system following the agentskills.io spec, MCP server integration, scheduled tasks, lifecycle hooks, and an evaluation harness for benchmarking agent behavior. The codebase demonstrates strong engineering practices: thorough unit testing (73%+ line coverage), clean module boundaries with leaf modules to prevent circular imports, and reusable patterns like typed event buses, retry decorators, and cancellation-safe streaming.
Responsively App is an Electron-based desktop application for responsive web development testing. It allows developers to preview a website across multiple device simulators simultaneously (phones, tablets, laptops) with synchronized interactions such as scrolling, clicking, and form input mirroring via BrowserSync. The application includes a built-in MCP (Model Context Protocol) server that exposes automation tools (navigate, screenshot, click, read_page) for AI coding agents like Claude Code and Cursor. It also provides design overlays, color blindness simulation, ruler/guide systems, and screenshot capture. The codebase is structured as a monorepo with three packages: the desktop app (Electron + React + Redux), a browser extension (Manifest V3) to open URLs in the app, and an MCP CLI package published to npm. The architecture follows a main-process/renderer split with IPC communication, extensive E2E tests (Playwright), and a device frame system with pixel-accurate bezel artwork generated from upstream sources.
This is a monorepo containing over 40 independent MCP (Model Context Protocol) servers for AWS services, hosted under awslabs/mcp. Each server exposes tools, resources, and prompts that allow AI agents (coding assistants, chatbots) to interact with AWS APIs, databases, documentation, and best practices. The architecture is a collection of standalone Python packages, each with its own entry point (server.py/main.py), tests, Dockerfile, and PyPI distribution. Servers are built using either FastMCP or the lower-level MCP Python SDK, with consistent patterns for read-only enforcement, error handling, AWS client creation, and user-agent tracking. The codebase demonstrates solid engineering with extensive test coverage, but suffers from significant copy-paste boilerplate across servers (healthcheck scripts, Context singletons, test_init.py) and inconsistent use of MCP frameworks. Notable high-craftsmanship components include the AWS API MCP server's custom CLI-to-IR parser and execution engine, the OpenAPI MCP server's pluggable authentication provider framework, and the DynamoDB MCP server's schema-driven code generation tool. The repository is actively maintained with frequent commits and issue triage, though some servers are marked as deprecated or in initial setup, and the announcement of a successor ('Agent Toolkit for AWS') suggests potential future migration.
Peekaboo is a mature macOS desktop automation toolkit that provides a CLI, Mac app, and MCP server for screen capture, UI element detection, input synthesis, window management, and AI agent integration. The architecture is modular: a Swift CLI (peekaboo) exposes ~33 commands for interaction, system control, and AI agent orchestration; a Mac app provides a menu bar interface, session management, and visualizer overlays; a bridge daemon (PeekabooBridge) enables secure remote automation via a versioned socket protocol with operation allowlisting. The MCP server exposes native automation tools to AI clients (Claude, Cursor, etc.) via stdio transport. The agent runtime (PeekabooAgentService) supports multiple AI providers (OpenAI, Anthropic, Ollama, etc.) with streaming, tool calling, and session persistence. The codebase is written in Swift 6.2 with strict concurrency, extensive unit/integration tests, and a focus on safety (fail-closed target validation, honest effect reporting). It is not a data research project; it is a production-grade application for macOS automation.
This repository is a Python library and framework for building and deploying multi-agent AI systems. It provides a structured way to define agents with specific roles, instructions, tools, and communication flows, orchestrated by an Agency. The framework extends the OpenAI Agents SDK with production-grade features: FastAPI web serving with streaming SSE endpoints, realtime voice support via WebSocket and Twilio, MCP (Model Context Protocol) server integration with OAuth authentication, persistent conversation state with load/save callbacks, system reminders for prompt engineering, input/output guardrails, and observability hooks for Langfuse/AgentOps. It also includes CLI tools for scaffolding agents and migrating from OpenAI Assistant configurations. Architecturally, the system is a library that can be used programmatically (create Agent and Agency instances, call get_response) or deployed as a FastAPI application via run_fastapi(). The core data flow involves user messages flowing through an agent's execution pipeline, which may invoke tools (built-in, MCP, OpenAPI, LangChain), hand off to other agents via SendMessage/Handoff, and stream responses back. The ThreadManager provides flat message storage with agent/caller metadata, supporting thread isolation and persistence. The tool system uses Pydantic models for schema generation and supports concurrency control. Key entry points include the Agent and Agency classes for programmatic use, run_fastapi() for web deployment, run_realtime() for voice, run_mcp() for MCP server, and CLI commands (create-agent-template, migrate-agent, import-tool). External dependencies include the OpenAI Agents SDK (pinned), FastAPI, Pydantic, and optional integrations like LiteLLM, LangChain, and MCP SDK. The codebase demonstrates strong engineering practices with extensive test coverage (unit, integration, live API), modular design, and reusable patterns such as the BaseTool class, ToolFactory adapter, ThreadManager message store, and OAuth flow for MCP. However, it has tight coupling to the OpenAI Agents SDK internals, requiring careful maintenance as the SDK evolves.
53AIHub is an open-source enterprise AI knowledge management and agent orchestration platform. The codebase reveals a multi-tenant SaaS/self-hosted system that allows organizations to build, deploy, and operate AI agents across multiple backends (OpenAI, Coze, Dify, Gemini, etc.), manage knowledge bases with full RAG pipelines (document parsing, chunking, embedding, graph extraction, wiki generation), record and transcribe meetings with insight extraction, and provide a unified search and chat interface. The architecture is a Go monolith backend using Gin and GORM, with a monorepo frontend comprising three React apps (admin console, end-user portal, and embeddable agent plugin) sharing a common business logic and component library. Data flows center on document ingestion pipelines, real-time agent relay with tool execution, and recording/speech processing. The system integrates with vector stores (Qdrant), Elasticsearch, Redis, and various LLM providers. While primarily an application, it contains substantial data processing pipelines characteristic of a data_research project, particularly the RAG and recording insight modules.
Commonly is an open-source multi-agent collaboration platform where humans and AI agents work together in shared workspaces called 'pods'. The system provides real-time messaging, task management, agent runtime (native, hosted, or bring-your-own), marketplace for agent skills, and integrations with external chat platforms like Discord, Slack, Telegram, and GroupMe. The architecture is a full-stack monolith with separate external agent bridge services and a Cloudflare Workers-based agent runtime. The backend (Node.js/Express) uses a dual-database design (MongoDB for primary data, PostgreSQL for messages/tasks) with automatic sync. The frontend is a React SPA with a v2 design system. The CLI allows local agents to participate via device-code auth and polling. The MCP server provides standard tools for agents to interact with the platform. The codebase is well-structured with extensive test coverage (unit, integration, E2E), Architecture Decision Records (ADRs), and reusable packages for types, integration SDK, and MCP tools. Notable craftsmanship includes pluggable storage drivers, deterministic agent wake targeting, and a test utility that replicates the production PostgreSQL schema to catch DDL drift.
This is the official Tableau MCP server, a TypeScript application that bridges AI assistants with Tableau Cloud, Tableau Server, and Tableau Desktop via the Model Context Protocol (MCP). It exposes a comprehensive set of tools covering content exploration (workbooks, views, datasources), administration (users, flows, extract refresh tasks, jobs, license reclamation), advanced analytics (Pulse metrics, insights, knowledge graph), and human-in-the-loop confirmation workflows. The server is available in three variants: a web variant running over HTTP with OAuth 2.1 support (PKCE, token rotation, client registration), a desktop variant communicating with the local Tableau Desktop Agent over stdio, and a combined variant. Its architecture is modular: a core server layer registers tools and prompts against an MCP SDK instance, tools delegate to a Zodios-based Tableau REST API client or the desktop agent API, and cross-cutting concerns like authentication, feature gates, telemetry, and session stores are provided through pluggable provider interfaces. The codebase demonstrates strong engineering practices with extensive unit/e2e tests, typed schemas via Zod, consistent error handling patterns, and a clear separation between web and desktop concerns.
This repository is an application: an MCP (Model Context Protocol) server paired with a Firefox browser extension that gives AI assistants like Claude Desktop read and management access to a user's real browser. The MCP server exposes nine tools over stdio — open/close/list/reorder/group tabs, search browsing history, extract a page's text and links, find/highlight text in a tab, and capture screenshots. Unlike generic web-automation MCP servers, it deliberately cannot click, modify pages, or run arbitrary scripts; page content access requires per-domain user consent and screenshots require per-tab authorization via Firefox's activeTab mechanism. Architecturally this is an Nx monorepo with three packages: mcp-server (Node 22, @modelcontextprotocol/sdk), firefox-extension (Manifest V2 WebExtension), and common (shared typed message protocol). The server communicates with the AI client over MCP stdio and with the extension over a loopback-only WebSocket (127.0.0.1 and ::1 on port 8089, or 0.0.0.0 in container mode). Every message is HMAC-SHA256 signed with a shared secret the extension generates; the extension connects outbound, so no inbound socket is exposed in the browser. Requests are matched to responses by correlation ID with timeouts, and the extension replies with resources (tabs, history, content, screenshots) over the same channel. The code is security-conscious and well-crafted: the dual-loopback binding handles IPv4/IPv6 localhost resolution differences, a capture-consent module mirrors the browser's activeTab grant lifetime to fail fast with actionable errors, the extension ships no runtime third-party dependencies, and it includes per-tool enable/disable toggles, a domain deny list, and a bounded in-browser audit log. The extension message handler has a solid Jest test suite covering consent revocation, denied domains, and disabled tools; the MCP server itself lacks unit tests. Minor debt: an unused, deprecated 'readline' npm dependency in mcp-server/package.json, an Nx version mismatch between root (22.x) and the subproject (20.x), and a few non-null assertions in the WebSocket layer. The repo is actively maintained with recent commits, a release workflow that builds DXT packages, and current dependency versions (TypeScript 5.8, Jest 30, MCP SDK 1.x, ws 8.x), so technical drift is low.
This is a monorepo for 'pi', an AI coding assistant platform. The repository contains multiple packages including a coding agent with a terminal UI (TUI), multi-provider AI integration (OpenAI, Anthropic, Google, Bedrock, Mistral, xAI, and more), a durable execution framework for conversation persistence, a protocol layer with strict CBOR encoding, and telemetry infrastructure. The coding agent supports interactive mode with session management, tool orchestration, subagents, and MCP (Model Context Protocol) integration. The architecture is a TypeScript monorepo using npm workspaces with packages for AI providers, coding agent, durable execution, protocol, telemetry, and TUI components. The AI package provides a unified interface across multiple providers with OAuth flows, model catalogs, and streaming support. The coding agent implements a TUI with themes, keybindings, and interactive components. The durable package provides event-sourced storage with SQLite and JSONL backends, session management, and compaction. The repository also includes evaluation infrastructure (Docker-based evals), benchmarks for storage and delta traversal, and telemetry for usage tracking. The codebase demonstrates strong engineering practices with clean abstractions, reusable components, and comprehensive test coverage, though some areas (durable harness, experimental services) are still evolving.
This repository is a structured, comprehensive AI engineering curriculum spanning 20 phases and over 500 lessons, from math foundations and ML fundamentals to deep learning, computer vision, NLP, speech, transformers, generative AI, reinforcement learning, LLMs from scratch, agent engineering, multi-agent systems, production infrastructure, and ethics/safety. It is an educational resource, not a production application or data research project. The code consists of self-contained pedagogical implementations (primarily Python, with some TypeScript, Rust, and Julia) that demonstrate core algorithms from scratch—such as attention mechanisms, BPE tokenization, diffusion models, RLHF, and agent loops—using synthetic data and deterministic stubs. Capstone projects include small web servers (Hono/TypeScript) and test harnesses. The curriculum is actively maintained with CI tooling, certification prep materials, and translation support. While the code is simplified for learning, it exhibits clean patterns, consistent lesson structure, and testing practices that make individual components (e.g., tool registries, state machines, evaluation harnesses) reusable in other contexts.
The google/adk-go repository is a production-grade, open-source Go framework for building, orchestrating, and deploying multi-agent AI systems. It provides a comprehensive set of abstractions including agents, workflows, tool integrations, model backends (Gemini, OpenAI, Apigee), session management, artifact storage, and human-in-the-loop (HITL) support. The architecture centers around a workflow engine that supports graph-based execution with nodes, edges, retry, concurrency, and dynamic branching. It also implements agent-to-agent (A2A) and Model Context Protocol (MCP) for inter-agent communication. The framework is designed to be deployed via REST API, WebSocket, SSE, or as a console application, and includes launchers for Google Cloud Run and Vertex AI Agent Engine. While primarily an application framework, it also includes telemetry and analytics plugins for observability. The codebase is well-structured with extensive unit and integration tests, but some internal packages and experimental features (e.g., OpenAI model adapter) indicate ongoing development and potential for interface drift.
EvoMap Evolver is a self-evolving AI agent engine built as a Node.js >=22.12 CLI daemon. It runs a genome-evolution loop: scans session memory/logs, extracts signals, selects 'Genes' (prompt/strategy patches) and 'Capsules' (validated skill bundles) from a local JSON/JSONL asset store, generates GEP-framed prompts for coding agents (Claude Code, Codex, Cursor, Gemini CLI, opencode, Kiro), executes those sessions, and records outcomes in a memory graph to adapt future selections. The system solidifies successful patterns into new reusable assets, auto-distills failure patterns and SKILL.md files into Genes/Capsules, and can open PRs or run validation sandboxes against candidate changes. It also ships an LLM proxy (Anthropic/OpenAI/Gemini/Vertex/Ollama routes with trace encryption and token-savings attribution), an ATP (Agent Transaction Protocol) marketplace for buying/completing agent tasks with auto-buyer and auto-deliver, a validation daemon for decentralized task verification, a local Web UI dashboard, and an experiment runner for A/B comparing evolution strategies. Hub synchronization (SSE event delivery, mailbox sync, OAuth device login) connects nodes to the evomap.ai network, but the system is designed to degrade gracefully into fully offline solo mode.
Greenshot is a mature, feature-rich Windows screenshot tool that goes far beyond basic capture. The codebase reveals a desktop application with a hybrid WinForms/WPF UI, a comprehensive plugin architecture for integrations (Jira, Confluence, Office, cloud storage, PDF, barcode), and an advanced recipe pipeline that automates capture workflows through a DAG of conditional steps. It includes a browser extension, native messaging proxy (C++), an MCP server for AI tool integration, and video recording via Windows Graphics Capture. The architecture is modular: a base library (Greenshot.Base) provides core services (capturing, threading, configuration, effects, file format handling, recipes, OAuth), while plugins register destinations, steps, and settings views. The test suite covers core flows, recipes, IPC, and threading, but UI and plugin integration tests are limited. Despite being a .NET Framework 4.8 codebase with some older dependencies (log4net, Dapplo), the engineering is solid—clean abstractions, async patterns, and a strong emphasis on thread safety and security (deserialization whitelist, IPC validation). The recipe pipeline and plugin system are particularly well-designed, though their complexity may present a learning curve.
ACI.dev is an open-source platform that provides a unified tool-calling interface for AI agents, enabling them to interact with over 600 third-party services via REST APIs or custom connectors. The backend is built with FastAPI, PostgreSQL (with pgvector for semantic search), SQLAlchemy, and Alembic for migrations, while the frontend is a Next.js 15 application with Shadcn UI, TanStack Query, and Zustand. The platform includes a full OAuth2 flow for service authentication, encrypted credential storage using AWS KMS, Stripe-based subscription billing, and a CLI for administrative tasks. It is designed as a multi-tenant SaaS with per-project quotas, agent management, and a playground for testing function calls. Architecturally, the system is modular: apps are defined as JSON schemas (app.json and functions.json) in a standardized format, and a generic execution engine dispatches calls to REST endpoints or local connector modules. The codebase demonstrates strong separation of concerns with CRUD modules, dependency injection, and middleware for rate limiting and logging. It also includes an evaluation pipeline for measuring search relevance (using synthetic intents and metrics like MRR), and a vector-based search for app and function discovery. While primarily an application, the repository contains research-oriented components such as the evaluation pipeline and embedding-based search, reflecting a hybrid nature focused on improving AI agent tool selection. The code is well-maintained and actively developed, with comprehensive test coverage (though some tests are skipped or TODO). The integration system is extensible, and several components (e.g., OAuth2 manager, encrypted SQL types, generic data table UI) are well-crafted and reusable. However, the codebase shows some technical debt in the form of hardcoded configurations, in-memory rate limiting, and app-specific OAuth2 overrides. Overall, it is a professional-grade platform with a high level of craftsmanship, suitable for production use and further extension.