Harness Manager

Harness Manager is a native Mac application that keeps an AI coding stack under control from one workspace. Instead of tracking tools, connections, and updates across many separate windows and tabs, users open a single app that shows what is installed, what is running, and what needs attention. From there they can discover harnesses, MCPs, and skills; install and update coding tools such as Claude Code, Codex, OpenCode, and Pi; inspect provider configuration, MCP servers, skills, and processes; and compare AI models across 31 ranking collections. It is built for developers and anyone who works with AI coding tools regularly and wants a clearer view of their setup. The app is free and open source, licensed under Apache 2.0. AI coding setups have grown quickly and unevenly. A typical developer may run several harnesses side by side, each with its own installation path, version history, and configuration. MCP servers add another layer of connections, skills add reusable instructions, and new models arrive constantly with different strengths for coding, design, reasoning, or local use. Keeping track of all of it often means keeping twenty browser tabs open, scrolling changelogs, and manually checking versions. It is easy to lose sight of which tools are actually installed, which ones have updates waiting, and where a broken installation or configuration problem is hiding. Harness Manager addresses that sprawl by putting the whole stack in one place so the next move — an update, a new skill, a model comparison — is visible rather than buried. The Discover screen is where new pieces of a workflow are found. Users browse harnesses, MCPs, and skills by popularity, moving from familiar tools like Claude Code and Codex to options they have not tried yet. Each entry is presented with install options, so the step from discovering a tool to having it on the machine is short. The app covers a broad catalog of harnesses, including Claude Code, Codex, Gemini CLI, Cursor, OpenCode, Warp, Antigravity, Antigravity IDE, T3 Code, Conductor, Superset, Paseo, cmux, Orca, Herdr, and Emdash. MCP servers and skills are discoverable through the same lens, which means connections and reusable instructions can be found and added without leaving the app. Popularity ordering gives a practical starting point when the ecosystem feels too large to survey manually. The workspace answers three questions at a glance: what is installed, what is running, and what needs an update. The update flow is deliberately transparent. Users see installed versions, compare them with available updates, and review the command that will run before deciding when it runs. Nothing is applied silently; the decision stays with the person at the keyboard. Harness Manager also automatically detects what is already installed on the Mac, checks versions and paths, and diagnoses broken installations or configuration issues. That makes it useful not only for adding new tools but for maintaining the ones already in daily use. When a tool fails to launch or a configuration drifts, the diagnostics surface the problem instead of leaving it to guesswork. The rest of a setup is shown in plain sight. Providers appear as local configuration signals, so it is clear how tools are pointed at the services they use. MCP servers are listed as the connections belonging to the tools, making it easier to see what each harness can reach. Skills are shown as reusable instructions, the building blocks that shape how a tool behaves. Running processes complete the picture by showing what is active and where it is running. Crucially, this information sits alongside the tools that use it rather than in a separate configuration file or terminal session. Grouping configuration with the tools it belongs to reduces the mental overhead of remembering which setting applies to which harness. Benchmarks and rankings turn model selection into a comparison rather than a hunt. The app presents 31 ranked collections covering general capability through creative and specialist work. General collections include Smartest, Coding, Agents, Fastest, Low latency, Cheapest, and Free. Development collections cover Design, UI components, Full-stack apps, Mobile apps, Tool calling, and Long-context reasoning. Reasoning and knowledge collections span Reasoning, Math, Science, Writing, Instruction following, RAG, and SQL and analysis. Deployment collections include Local, Open-source, Small and fast, Long context, Vision, and Uncensored. Creative and specialist collections cover Data visualization, SVG, Game development, 3D, and Roleplay. Seven comparison metrics are available, including intelligence, coding, agentic performance, design, speed, latency, and context. Rankings data is provided by Modelgrep. The app is explicit that rankings are a starting point and results depend on the task. The Harness Briefing keeps the ecosystem readable without opening twenty tabs. It gathers publisher news, community finds, and official releases into one native page. Sources include OpenAI, Google Developers, Simon Willison, Hacker News, and project releases. Every story links back to its source, so readers can verify and read further, and articles are read inside the app. The intent is to go beyond the changelog — new harnesses, useful ideas, and the story behind a release — so that staying current is a matter of checking one screen rather than monitoring many feeds. Getting started takes three steps. First, download, drag, and open: Harness Manager moves to Applications and the workspace opens, with no build tools required. Second, see the existing stack: the app finds supported tools and configuration already on the Mac and brings them into one view. Third, make the next move: review an update, discover a skill, or compare models for the next project. The app requires macOS 14 or later and runs on both Apple silicon and Intel Macs. It is distributed as an early preview build that is not yet notarized, so macOS may require approval in Privacy & Security on first launch. The project is free to use, inspect, and modify under the Apache 2.0 license, with source available on GitHub. Benefits follow directly from consolidation. Users spend less time on setup management and more time building, because the state of the stack — installations, versions, connections, skills, and processes — is visible in one place. Updates become a deliberate decision rather than a surprise, since the command is reviewed before it runs. Diagnostics reduce the time lost to broken installations and configuration issues. Model comparison shortens the path from a new task to a sensible starting model, and the briefing replaces scattered reading with a single sourced feed. For teams and individuals who depend on AI coding tools every day, the outcome is a stack that stays sorted and a workflow that keeps moving. Use cases span the daily rhythm of working with AI coding tools. A developer setting up a new Mac can let Harness Manager detect existing installations and fill in what is missing from the Discover screen. Someone maintaining several harnesses can check installed versions, compare available updates, and review the command before applying it. A user troubleshooting a tool that will not start can look for broken installations or configuration issues and inspect provider configuration, MCP servers, and processes. When starting a new project, they can browse ranking collections to shortlist a coding or reasoning model. And when they want to keep up with the ecosystem, they can read the briefing's sourced stories in the app. The app is aimed at developers and AI tool users who run coding harnesses on a Mac and want their environment legible rather than scattered. It supports a catalog that includes Claude Code, Codex, Gemini CLI, Cursor, OpenCode, Warp, Antigravity, Antigravity IDE, T3 Code, Conductor, Superset, Paseo, cmux, Orca, Herdr, and Emdash. Requirements are macOS 14 or later on Apple silicon or Intel hardware. Harness Manager is free and open source under Apache 2.0, available as a direct download from GitHub, and supported by an invitation to star the project if it helps keep a stack moving. It is also listed on Product Hunt. Harness Manager's value proposition is simple: one workspace for an AI coding stack. It combines discovery of harnesses, MCPs, and skills with update control, configuration visibility, diagnostics, model rankings, and an ecosystem briefing, all in a native Mac app that is free and open source. The result is less setup to manage and more space to build.