#5
AnyPS5 is an open-source tool that automatically ports PS5 executables to Linux and Windows by converting them to the target system's native format via a relinker, without emulation or a separate runtime process. It also provides implementations of PS5 system PRX libraries suitable for dynamic linking, and includes a shader recompiler that produces SPIR-V output validated via SPIRV-Tools. A web-based converter is available, keeping files local to the user's device, and a relinker-only build supports macOS including Apple Silicon. The project tracks per-game compatibility progress through a community-maintained blocker map built from public GitHub data, with test reports organized in dedicated discussion threads. SDL-mapped controllers, keyboard, and mouse input are supported with configurable INI-based mappings. The project is actively developed but still maturing, with library and shader coverage tracked via live progress badges. It is notable for its no-emulation approach to PS5 compatibility, making it a technically ambitious and trending project in the game preservation and interoperability space. Licensed under GPL-2.0 only.
MCPAppslocal-firston-device
★ 43,396 total+12,764 this week
#13
Magpie is an open-source local gateway application that acts as a universal model-switching layer for AI coding agents, allowing developers to run any supported agent (such as Claude Code, Codex, Gemini CLI, or OpenCode) against any AI provider or model from a single unified interface. The core problem it solves is fragmentation: each agent stores its model configuration, API keys, and base URLs in its own proprietary format, and subscriptions are siloed to one agent. Magpie abstracts all of this behind a local gateway running at 127.0.0.1:3425, which speaks OpenAI Chat, OpenAI Responses, Anthropic Messages, and Gemini protocols simultaneously, translating between them with full support for streaming, tool calls, and reasoning. Users can switch models for any agent from a menu bar, window, TUI, web UI, or CLI, and the agent's config file is rewritten atomically while preserving comments, ordering, and formatting. A standout feature is routing groups, which bundle multiple models or accounts under one logical name and support five routing strategies — smart (quota-aware), order (fallback), rotate (round-robin), usage (least-used-first), and pace (time-balanced) — enabling seamless failover when quotas are exhausted without the agent ever seeing an error. Intent routing adds an additional layer where a small model reads each conversation turn and dispatches it to the appropriate model based on complexity, routing heavy tasks to powerful models and trivial queries to cheaper ones. Magpie supports an extensive list of over 60 named agents and dozens of provider presets including Anthropic, OpenAI, Google Gemini, Vertex AI, Amazon Bedrock, DeepSeek, Kimi, Mistral, Groq, xAI, Ollama, and many others, with model metadata sourced from models.dev and refreshed automatically. Subscription accounts (Claude, ChatGPT, GitHub Copilot, Gemini Code Assist, Grok SuperGrok, and others) are treated as first-class providers, with daily warm-up scheduling to maximize 5-hour usage windows. A plugin system built on Bun allows community-authored or custom plugins to add new auth flows, model sources, and gateway middleware. The project runs on macOS, Windows, Linux, Docker, and Termux, is MIT-licensed, and appears to be gaining significant traction as a trending tool in the rapidly growing AI coding agent ecosystem, particularly for power
MCPAgentsServingcodinglocal-firstworkflow-automation
★ 16,262 total+6,637 this week
#29
Context Mode is an MCP (Model Context Protocol) server that addresses what its authors call "the other half of the context problem" in AI-assisted coding workflows. The core issue it targets is context window bloat: every MCP tool call dumps raw data into an LLM's context window — a Playwright snapshot costs 56 KB, twenty GitHub issues cost 59 KB, and a single access log can consume 45 KB. After roughly 30 minutes of work, 40% of the context can be lost, and when agents compact conversations to free space, they forget which files they were editing, what tasks are in progress, and what the user last requested. On top of this, agents waste output tokens on filler, pleasantries, and verbose explanations, burning context from both input and output sides. Context Mode tackles all four facets of this problem through a unified approach. First, **Context Saving**: sandbox tools keep raw data out of the context window, reducing 315 KB to 5.4 KB — a claimed 98% reduction. Second, **Session Continuity**: every file edit, git operation, task, error, and user decision is tracked in SQLite, with events indexed into FTS5 and retrieved on demand via BM25 search rather than being dumped back into context; this lets the model resume exactly where it left off, while sessions started without `--continue` get a clean slate. Third, **Think in Code**: the paradigm that the LLM should program analyses rather than compute them manually — for example, instead of reading 50 files into context to count functions, the agent writes a script that computes and `console.log()`s only the result, replacing many tool calls and saving up to 100x context. This paradigm is mandatory across all 17 supported clients plus the OpenClaw gateway integration. Fourth, **No prose-style enforcement**: Context Mode controls where data goes but deliberately never dictates how the model phrases its final answer, citing research (attributed to Moonshot AI on kimi-k2.5) that aggressive brevity prompts degrade coding and reasoning benchmarks; formatting and brevity remain the user's or model's choice via CLAUDE.md or AGENTS.md. The project appears mature and feature-rich. For Claude Code, it offers a fully automatic plugin marketplace install, with a SessionStart hook injecting routing instructions at runtime without writing files to the project. The plugin registers hooks (PreToolUse, PostToolUse, UserPromptSubmit, PreCompact, SessionStart, Stop) and 11 MCP tools: six sandbox tools (ctx_batch_execute, ctx_execute, ctx_execute_file, ctx_index, ctx_search, ctx_fetch_and_index) and five meta-tools (ctx_stats, ctx_doctor, ctx_upgrade, ctx_purge, ctx_insight). Slash commands provide diagnostics, indexing, searching, upgrading, purging, context-savings statistics, and a hosted Insight dashboard offering org-level analytics for AI engineering teams. An optional status line shows real-time savings. The README documents installs for multiple platforms grouped by complexity — hook-capable platforms get automatic routing enforcement while others need a one-time routing file copy — including detailed Gemini CLI setup via a single settings.json registering the MCP server and four hooks (BeforeTool, AfterTool, PreCompress, SessionStart), with carefully scoped matchers targeting only high-output tools. It requires Node.js 22.5+ or Bun, is distributed on npm, and is licensed under the Elastic License v2 (ELv2). Signs of momentum and notability are strong: the badges claim adoption across teams at major companies, a Discord community, a YouTube demo, and a prominent Hacker News ranking (#1 with 570+ points). The project emphasizes measurable, real-time context savings and positions itself as a disciplined alternative to brevity-prompt approaches. Notably, the long list of company logos appears to be marketing badges rather than verified endorsements, so the adoption claims should be treated cautiously. Overall, Context Mode is a polished, actively maint
MCPToolsServingcodingworkflow-automation
★ 30,828 total+4,111 this week
#33NEWPartial week
SolidCraft is an open-source parametric 3D CAD application built entirely in Rust, developed by Tangency Labs. It targets engineers, makers, and designers who want a modern, cross-platform solid modeling tool that runs natively on Windows, macOS, and Linux, or directly in the browser with no installation required. The core paradigm is a parametric feature timeline: users sketch geometry, apply constraints and dimensions driven by named parameters, then build up solids through features like extrude, revolve, sweep, loft, fillet, chamfer, shell, and draft. Changing a single dimension causes the entire model to rebuild automatically, which is the hallmark of parametric CAD. A standout and highly notable feature is a built-in MCP (Model Context Protocol) server that allows AI agents to drive the modeling process programmatically, enabling fully automated design workflows. A showcase example is SELENE, a scientific rover with 3,323 editable timeline steps built entirely through the MCP server by a community contributor, demonstrating the system's ambition and real-world capability. The constraint solver is custom-built, combining geometric construction techniques (loci intersections, rigid-cluster decomposition based on Owen 1991 and Fudos–Hoffmann 1997) with Levenberg–Marquardt numerical solving, and reports remaining degrees of freedom via Jacobian rank analysis. Sketch tools are comprehensive, covering lines, rectangles, circles, arcs, polygons, slots, ellipses, splines, conics, and text, with full support for construction geometry, trimming, fillets, chamfers, and associative offsets. Profile detection is automatic, including regions split by crossing lines and holes within outlines. The application also supports assemblies with placed components, DXF export of current views with hidden-line handling, and material definitions with mass properties. Distribution is broad: Windows MSI (x64, ARM64, x86), portable zip, WinGet, macOS universal DMG, Linux AppImage, .deb, .rpm, Flatpak, and Snap packages are all available, plus a web app. Maturity is described honestly as "early but usable for simple parts," with 75 of 76 public reference parts passing volumetric, area, center-of-mass, and bounding-box validation against analytic or FreeCAD-computed references. A roadmap, feature list, oracle test suite, and contributing guide are all present, signaling an organized and transparent open-source project. The combination of Rust performance, browser deployment, AI-agent integration via MCP, and a rigorous parametric engine makes SolidCraft a notably ambitious and trending entry in the open-source CAD space, especially as AI-assisted design workflows gain momentum.
MCPAppsAgentsworkflow-automationlocal-first
★ 5,951 total+3,648 this week
#46
OmniRoute is a free, open-source AI gateway designed to unify access to a vast array of AI providers through a single endpoint, allowing developers to "never stop coding" by avoiding rate limits and provider lock-in. The project's core proposition is routing every AI tool — including Claude Code, Codex, Cursor, Cline, Copilot, and Antigravity — into free Claude, GPT, and Gemini models with automatic fallback. As of the documented versions, OmniRoute catalogs 375 AI providers (up from 352 in prior releases) and exposes 1,475 unique chat model IDs, with more than 150 providers marked as free in its catalog. A central headline feature is the aggregation of free-tier token quotas: OmniRoute estimates roughly 1.62 billion free tokens per month across separately eligible third-party free tiers, rising to approximately 2.22 billion in the first month when signup credits are included. Importantly, the project is transparent that these tokens are not granted by OmniRoute itself — users must bring their own provider accounts and credentials, and each provider controls its own limits, availability, and terms. The token math is derived from 468 cataloged free-tier entries across 34 recurring pool keys, computed from 17 pools with a published positive monthly budget plus five per-model Groq caps, deduplicated so shared pools are counted only once. Quotas gated behind regional identity verification (such as ModelScope, adding ~6M tokens) are listed separately and excluded from the headline, and 12 providers are flagged as "avoid" in a terms-risk catalog so users can make informed decisions. These figures are re-audited every two weeks and can move in either direction, with live usage visible on a dashboard at `/dashboard/free-tiers`.
Beyond aggregation, OmniRoute offers 20 routing strategies and several notable features. Its RTK + Caveman "stacked compression" technology reportedly saves 15–95% of tokens (approximately 89% on average) to help users stay within limits. The gateway includes a Modality Bridge supporting vision, audio, and video; a Radar free catalog (opt-in); quota-aware scheduling via a "Quota-Share" mechanism; and live quota telemetry. Newer roadmap items heading toward a v3.9.0 LTS release include a headless mode (`serve --headless`), LTS "rail" infrastructure with release channels, and a model discovery API accessible at `/v1/models/query`. The project emphasizes being private and local-first, and highlights full CLI support alongside A2A (agent-to-agent) and MCP (Model Context Protocol) integrations — relevant to the "a2a" query, OmniRoute advertises combined CLI, A2A, and MCP capabilities, positioning it as interoperable with agent ecosystems and compatible coding agents. "Combos" are described as a flagship feature of the platform.
In terms of maturity and distribution, OmniRoute is actively maintained and distributed across multiple channels: it is published on npm (as the `omniroute` package), available on Docker Hub, and offered as an Electron desktop application with download counts tracked on GitHub. It is MIT-licensed and localized into 67 languages, with translations for Portuguese (Brazil and Portugal), Spanish, French, Italian, German, Dutch, Russian, Ukrainian, Polish, Czech, Slovak, Romanian, Hungarian, Bulgarian, Danish, Finnish, Norwegian, Swedish, and many others. The project shows strong community traction: it features on Trendshift (repository #23589), tracks star history, and claims over 900 contributors. Community support is organized through Discord, Telegram, and WhatsApp (both global and Brazil-specific groups), plus a dedicated website (omniroute.online) and portal. The maintainer, Diego Souza, promotes updates via LinkedIn and GitHub. OmniRoute runs "anywhere" across platforms, and provides extensive documentation including a free-tiers methodology reference and a roadmap. The project is notable and trending because it addresses a
MCPServingToolslocal-firstmultimodalworkflow-automation
★ 78,208 total+2,827 this week
#60
The `cursor/plugins` repository is the official collection of agent plugins for Cursor, the AI-powered code editor, designed to extend its capabilities across popular developer tools, frameworks, SaaS products, and productivity platforms. Each plugin lives as a standalone directory at the repository root and is defined by a `.cursor-plugin/plugin.json` manifest, making the architecture modular and easy to extend. The plugin catalog is already impressively broad, spanning four major categories: Utilities, Developer Tools, Productivity, and Integrations. On the developer tooling side, plugins cover sophisticated agentic workflows such as parallel cloud agent orchestration (`orchestrate`), deep security and code-quality branch audits (`thermos`), iterative self-referential AI loops (`ralph-loop`), CLI design patterns for coding agents (`cli-for-agent`), and continual learning via transcript-driven memory updates. There are also plugins for scaffolding new plugins (`create-plugin`), consulting stronger models before major decisions (`advisor`), and building on the Cursor Origin API and TypeScript SDK. The productivity and integration plugins are extensive, covering the full Google Workspace suite (Gmail, Drive, Calendar, Docs, Sheets, Slides), CRM platforms (Salesforce, HubSpot, Attio), sales and marketing tools (Outreach, Klaviyo, Customer.io, Amplemarket, Brevo), recruiting tools (Ashby, Workable, Juicebox, Upwork), meeting intelligence (Gong, Fireflies, Otter.ai, Fathom, Zoom, Circleback), financial platforms (Xero, Brex, Interactive Brokers, Webull, S&P Global), advertising (Google Ads, Meta Ads, X Ads), SEO tools (Semrush, Ahrefs, Similarweb), and many more. The repository is clearly in active, early-stage growth — a rapidly expanding ecosystem rather than a mature, stable release — with contributions from both the Cursor team and community authors like Eric Zakariasson, Lauren Tan, and Dylan Gattey. Its trending appeal lies in the fact that it represents Cursor's push to become a universal AI agent platform, not just a coding assistant, by enabling agents to interact
MCPAgentsToolsworkflow-automationcoding
★ 12,821 total+1,896 this week
#74
KytyPS5 is a free and open-source PlayStation 5 translation layer written in C++, targeting Windows x64 and Linux x64 as primary platforms, with experimental macOS support via Rosetta 2 on Apple Silicon. The project is a heavily modified fork of the earlier Kyty emulator by InoriRus, now evolved into a dedicated PS5 compatibility layer. Rather than relying on external low-level emulation modules, KytyPS5 implements its own internal pipeline, including shader decoding and recompilation, GPU command processing, and a Vulkan 1.3 host backend. The PS5 GPU architecture is based on AMD RDNA 2, and the project references AMD's official RDNA 2 Instruction Set Architecture Reference Guide (document 70648) for shader instruction encoding. On macOS, Vulkan support is provided through the bundled MoltenVK library. The project can already boot a notable range of 2D and 3D titles, including games built on Unreal Engine 4/5, Unity, and custom engines, with verified in-game screenshots of titles such as Astro Bot, Demon's Souls, Neptunia ReVerse, SILENT HILL: The Short Message, Saros, and UFC 6. A public compatibility list is maintained at kytyps5.github.io, and weekly development updates are posted in GitHub Discussions. The build system uses CMake with Ninja and requires Clang (clang-cl on Windows, clang/clang++ on Linux), explicitly rejecting the MSVC compiler. Qt 6 is required for the launcher, while a headless emulator build is also supported on Linux without Qt. FFmpeg is statically linked from a pinned KytyPS5 FFmpeg core fork, providing VP9 and WebM support without requiring system FFmpeg packages. NixOS and Fedora/Fedora Atomic variants (Bazzite, Bluefin, Silverblue) receive dedicated build instructions, with Distrobox recommended for immutable Fedora systems. The project enforces strict contribution guidelines: pull requests must target a specific game bug, and unfocused or architectural-only PRs are automatically closed. Code formatting is enforced via clang-format through a pre-commit Git hook. Licensed under GPL-2.0, KytyPS5 is notable as one of the few actively developed, open-source PS5 translation layers with cross-platform CI builds and a growing game compatibility record, making it
MCPAppsmultimodallocal-first
★ 9,091 total+1,485 this week
#87
Open SEO MCP Skills is a free, MIT-licensed SEO and GEO (Generative Engine Optimization) toolkit that connects Claude to real SEO data via the Model Context Protocol (MCP). Unlike typical "open source SEO tools" that wrap DataForSEO with markup fees, this project pulls rankings from actual Google Search Console, traffic from real GA4 data, and keyword volumes from Google Ads—eliminating estimates. It includes eight ready-made skills: site audits, keyword research, rank tracking, competitor gap analysis, backlink checks, AI visibility reporting, content briefs, and SEO-vs-ads overlap analysis. The AI visibility skill is particularly notable, tracking actual referral traffic from ChatGPT, Perplexity, Claude, and Gemini via GA4. Setup is minimal—three CLI commands or a manual connector URL in Claude settings. The project is early-stage but functional, positioning itself as the only SEO MCP server running on verified first-party data rather than scraped estimates. It's trending at the intersection of the MCP ecosystem boom and growing interest in AI-era SEO tooling, with no subscription cost and no per-call markup, making it notable for developers and SEO practitioners adopting Claude Code workflows.
MCPToolsAppsworkflow-automationbrowser-use
★ 6,069 total+1,238 this week
#97
The `anthropics/knowledge-work-plugins` repository is an open-source collection of 11 role-specific plugins that extend Claude's capabilities within Claude Cowork and Claude Code environments. Each plugin bundles domain expertise, slash commands, connectors, and sub-agents tailored to a specific job function — covering sales, marketing, legal, finance, data analysis, product management, customer support, productivity, enterprise search, bio-research, and plugin management itself. Plugins connect Claude to real-world tools like Slack, HubSpot, Snowflake, Jira, Notion, and Figma via MCP servers, enabling finished, professional work output rather than just conversational assistance. The architecture is entirely file-based (markdown and JSON), requiring no code or infrastructure, making customization accessible to non-engineers. Teams can swap connectors, inject company-specific terminology, and modify workflows to make Claude behave as if purpose-built for their organization. The project is early-stage but notable for its breadth of integrations, its no-code extensibility model, and its positioning as a marketplace-style ecosystem for enterprise AI workflows. Its GitHub trending status likely reflects growing enterprise interest in structured, role-aware AI agents built on top of frontier models.
MCPAgentsToolsworkflow-automation
★ 30,257 total+1,105 this week
#106NEWPartial week
MailCheap is an open-source newsletter platform built around a pay-per-email pricing model, meaning users are charged only for the emails they actually send rather than for the total number of subscribers on their list — a direct contrast to platforms like beehiiv, Substack, and Mailchimp that typically charge based on subscriber count. The project is authored by marclou and licensed under AGPL-3.0, positioning it as a transparent, self-hostable alternative to mainstream newsletter services. The entire application — both the cloud-hosted version and the self-hosted version — runs from this single repository, with MailCheap Cloud simply being the same codebase deployed with a `MAILCHEAP_CLOUD=true` flag that enables open sign-up, plan management, and Stripe billing.
A standout feature of MailCheap is its deep integration with AI agents, which is directly relevant to the ai-agents trend query. The platform exposes a Model Context Protocol (MCP) server that allows AI agents such as ChatGPT, Claude, Grok Bot, Claude Code, Codex, and Cursor to perform virtually every action available in the dashboard — writing, reviewing, and sending newsletters — through either OAuth or API key authentication. Critically, the MCP server marks sending as an irreversible action, prompting agents to seek user confirmation before dispatching emails, which reflects a thoughtful approach to agentic safety and human-in-the-loop design. This makes MailCheap one of the relatively few newsletter tools explicitly architected for AI agent workflows from the ground up.
The campaign workflow supports writing via a rich editor with slash commands, images, and layout boxes, or via Markdown and raw HTML submitted through the API or MCP. A review step checks for common issues before sending — subject line, preview text, unsubscribe link, Gmail's 100 KB clip threshold, postal address, and sender identity. Campaigns can be sent immediately, scheduled, or rolled out in a warm-up pattern where the most engaged readers receive the email first, with broader delivery following only if bounce and complaint rates remain acceptable.
Deliverability is treated as a first-class concern with multiple layered protections. These include automatic list warm-up, list cleanup that re-engages dormant readers before removing them, contact cleaning that filters invalid, misspelled, spam trap, role-based, throwaway, and duplicate addresses at import time, and a server-wide suppression list that stores only keyed hashes of problematic addresses. Hard bounce and spam complaint thresholds automatically pause or block sending accounts, and the platform integrates with Google Postmaster Tools to monitor Gmail-specific spam rates on a daily basis. Every email includes RFC 8058 one-click unsubscribe and a signed unsubscribe link, with confirmed opt-in enabled by default.
The platform supports importing subscriber lists from beehiiv, Substack, Mailchimp, Kit, MailerLite, Buttondown, Ghost, and EmailOctopus, preserving subscriber status and signup dates. Each newsletter gets a public-facing website with a signup form, issue archive, author page, share images, and a sitemap, hosted either on a free MailCheap Cloud subdomain or on the user's own server. Custom domain sending is supported with guided DKIM, return path, and DMARC configuration.
For self-hosting, the stack requires MongoDB for data storage, Amazon SES or PostShiba for email delivery, Upstash QStash for background job scheduling, and a Node.js 22+ host such as Vercel. Optional services include Resend for transactional sign-in emails, Upstash Redis for shared rate limiting across serverless instances, Amazon S3 with CloudFront or Vercel Blob for image uploads, and Google Postmaster Tools for deliverability monitoring. A one-click Vercel deploy button is provided, and setup is estimated at roughly one hour, mostly waiting on DNS propagation and provider approval
MCPAppsAgentsworkflow-automationlocal-firstsecurity
★ 1,237 total+999 this week
#110
OpenSEO is an open-source, pay-as-you-go alternative to expensive SEO platforms like Semrush and Ahrefs, designed for individuals and AI agents. It covers core SEO workflows including keyword research, rank tracking, competitor insights, backlink analysis, site audits, and AI visibility monitoring. A standout feature is its MCP (Model Context Protocol) server, which allows AI agents such as Claude Code, OpenClaw, and Hermes to directly consume SEO data through reusable "Agent Skills" — making it notable in the growing AI-agent tooling space. Rather than bundling data costs into a subscription, OpenSEO requires users to bring their own DataForSEO API key, paying only for actual usage. Self-hosting is supported via Docker or Cloudflare (recommended for multi-device/team use), and a hosted version is available at $10/month. The project appears early-stage but actively developed, with community channels on Discord and X. Its combination of open-source flexibility, AI-agent integration, and cost transparency positions it as a trending tool at the intersection of SEO and agentic AI workflows.
MCPToolsAppsworkflow-automationbrowser-use
★ 24,053 total+919 this week