Hadron MCP server
Commit Nov 13, 2026Orchestrate multi-provider LLM swarms in an isolated Rust-based workspace with automated git worktree management and merge gating.
Also known as s0lda/hadron
- Stars
- 29
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Nov 13, 2026
- License
- Apache-2.0
What is Hadron MCP?
Hadron orchestrates multi-provider LLM swarms in an isolated Rust-based workspace with automated git worktree management and merge gating.
Summary generated from the project's README
About Hadron MCP server
Orchestrating multi-provider LLM swarms in an isolated Rust-based workspace is what Hadron enables for developers managing complex agentic workflows. It allows users to combine frontier models like Anthropic Claude, OpenAI, and xAI Grok with local providers such as Ollama or OpenRouter within a single unified environment. The system uses a hub-and-spoke orchestration model where agents, referred to as Quarks, collaborate through @mentions over a zero-CPU filesystem event bus.
Developers use this desktop workspace to run concurrent agents that work safely within isolated git worktrees. To ensure code quality, an automated Merge Gate rebases changes onto the main branch and executes test suites before any code is committed. The platform includes specialized tools like the Power Tools Suite for LSP code intelligence, headless browser testing, and AST Edit-by-Hash to prevent line-drift corruption. It supports various operating systems including Linux, macOS, and Windows through native process sandboxing.
What you can do with it
- 1Collaborative Multi-Model SwarmsAssign tasks to different AI models simultaneously within a single chat thread using mentions.
- 2Automated Code ValidationEnsure all agent-generated code passes tests before it is merged into your primary branch.
- 3Safe Agentic ExecutionRun AI agents in isolated environments to prevent them from corrupting your local files or repository.
- 4Efficient Token ManagementReduce the cost of running multiple agents by utilizing resident prompt caching and context isolation.
Written from the project's README. Every use case is checked against a quote from it.
Automated checks
Ran Nov 14, 2026Full report →Not verified yet. A server needs a score of 80, no serious finding, and enough data for us to check. How checks work · What to fix for more points
- Source repository linked and reachables0lda/hadron is public and reachablegithub.com/s0lda/hadron ↗Pass
- Package names this repository as its sourceNot checked
- Build provenance (Sigstore attestation)Not checked
- Open source license (OSI)Apache-2.0spdx.org/licenses/Apache-2.0.html ↗Pass
- Commit or release in the last yearLast commit or release 2026-11-13github.com/s0lda/hadron ↗Pass
- OpenSSF ScorecardNot checked
- Package name not close to a popular packageNot checked
- No critical advisory in the package or its dependenciesNot checked
- No hidden instructions to the model0 tool descriptions and the README read; no override phrases, credential paths or hidden charactersPass
- Destructive tools are labelledNot checked
- Remote endpoint answers the MCP handshakeNot checked
README
View on GitHub🌌 Hadron
The Native Desktop Workspace for Multi-Provider LLM Swarms
Orchestrate Frontier LLMs (Claude, Antigravity, OpenAI, Grok) and Local/OpenRouter Agents in Isolated Git Worktrees with Automated Merge Gating
⚡ What is Hadron?
Hadron is a native, GPU-accelerated desktop workspace built in Rust for orchestrated multi-provider LLM swarms. It brings frontier models (Anthropic Claude, Google Antigravity, OpenAI, xAI Grok) together with OpenRouter, local Ollama / vLLM, and any custom ACP or coding CLI agent in a unified room where agents collaborate, exchange context, and delegate tasks using @mentions.
Behind the interface, agents (called Quarks) execute concurrently over a zero-CPU filesystem event bus (field.jsonl), work safely in isolated git worktrees, and submit code changes through an automated Merge Gate that runs tests before landing anything on main.
💡 Key Superpowers
- 💬 Universal Multi-Model Swarms: Mix frontier models (
@claude,@agy,@openai,@grok), OpenRouter, local@ollama, and custom CLI agents in one thread via Hub-and-Spoke@mentionorchestration over a zero-CPUfield.jsonlbus. - 🛡️ Collision-Proof Worktrees & Merge Gate: Quarks work in isolated git worktree branches. The automated Merge Gate rebases onto
mainand runs your real test suite before clean code lands — with Hadron Lab concurrent merge gates, speculative batch evaluation, and docs/asset fast-paths. - 💰 Token-Efficient Swarm Economics: Cuts multi-agent input token spend by 80–90% via resident prompt caching (
cache_read), Hub-and-Spoke context isolation (preventing $O(N^2)$ sibling log broadcasts),ContextGovernoroutput elision, lazy pointer-based Nucleus memory, and live 4-way USD/quota telemetry. - 🔒 Cross-Platform Process Sandboxing: Zero-overhead native OS containment across Linux/WSL (Bubblewrap), macOS (Apple Seatbelt), and Windows (PowerShell & Job Objects) with worktree write confinement, credential scrubbing, and interactive chat badge toggles.
- 🛠️ AST Edit-by-Hash (
blake3) & 50+ MCP Forge Tools: Eliminates LLM line-drift corruption via content-addressed AST precision edits, plus isolated PTY terminals, headless browser testing, LSP code intelligence, automatedgit bisect, and flamegraph profiling (Power Tools Suite). - ⚡ 120 FPS GPUI Desktop, Headless CI &
/absorbMemory: Native Rust workspace with interactive File Tree CRUD, multi-tab PTY terminals, Git & Wave DAG visualizers, standalonehadron run/hadron cirunners, 29 bundled engineering skills, and/absorbto import rules from.claude/,.cursor/,.windsurf/, and.agents/.
🚀 Quick Start
Prerequisites
- OS: Linux (X11/Wayland), macOS, or Windows (native MSVC or WSL2).
- LLM Access: Any ACP agent (e.g. Claude Code), Antigravity, local Ollama, or API key for OpenAI / OpenRouter / Anthropic.
1. Install Hadron
Quick Install (Prebuilt Binaries — Recommended)
Linux & macOS:
curl -fsSL https://raw.githubusercontent.com/s0lda/hadron/main/install.sh | sh
Detects Linux and macOS architectures, validates SHA256 checksums, and installs all three binaries (hadron, hadron-gluon, hadron-forge-mcp) into ~/.local/bin.
Windows (PowerShell):
irm https://raw.githubusercontent.com/s0lda/hadron/main/install.ps1 | iex
Installs Hadron to $env:LOCALAPPDATA\Programs\Hadron, verifies SHA256 checksums, configures user PATH, and sets up Start Menu / Desktop shortcuts with the custom Hadron application icon.
Build from Source (Cargo)
Requires Rust toolchain (Rust 1.93+, edition 2021).
cargo install --locked --git https://github.com/s0lda/hadron.git hadron
2. Launch in Any Repo
# Open or create a swarm workspace in your project directory
cd ~/dev/my_project && hadron
# Or initialize Git + auto-discover local CLI agents & API keys from the terminal
hadron init
# Or execute headlessly in CI / terminal batch mode
hadron run "Implement user authentication with tests"
hadron ci --plan .hadron/docs/plans/<plan_name>
📚 Documentation & Deep Dives
Explore Hadron's mechanics, architecture, and developer guides in the /docs directory:
| Section | Description | Links |
|---|---|---|
| 📜 Changelog | Release history & version notes | Changelog |
| 🕹️ Commands | Full interactive chat commands reference | Commands Reference |
| 🏛️ Architecture | 2-Tier Daemon/GUI split & Swarm Event Loop | Decoupled Architecture · System Diagrams |
| ⚛️ Concepts & Physics | Particle Physics Mental Model & Glossary | Physics Metaphor · Vocabulary |
| 🛠️ Hadron Forge | AST Edit-by-Hash & Developer Power Tools | Edit-by-Hash Engine · Power Tools Suite |
| 🌿 Development | Building from Source & Git Worktrees | Building from Source · Git Worktrees |
🔌 Provider & Model Ecosystem
Hadron connects an expanding ecosystem of AI agents, local models, cloud APIs, and terminal tools into a single orchestrated workspace. Any binary, endpoint, or script matching supported protocol interfaces can be seated directly into your swarm:
- 🤖 ACP (Agent Client Protocol): Resident agent processes with automatic session resumption, model selection, and tool execution.
- 🌉 Custom SDK Bridges: Embedded adapters for SDK-based agent runtimes (such as Google Antigravity).
- ☁️ Local & Cloud HTTP Endpoints: Zero-config connection to local model servers (Ollama, LM Studio) and cloud model aggregators or APIs (OpenAI, OpenRouter with 400+ models, or any OpenAI-compatible endpoint).
- 💻 CLI & PTY Transports: Stdio wrappers (
CliSpec) for terminal scripts and interactive PTY terminal seats alongside LLM agents in the chat room.
⌨️ Essential Keyboard Shortcuts
| Shortcut | Action |
|---|---|
Ctrl+Tab / Ctrl+\ |
Toggle focus between Chat Input and PTY Terminal |
Ctrl+Shift+T / Ctrl+Shift+W |
New / Close Terminal Tab |
Ctrl+PageDown / Ctrl+PageUp |
Next / Previous Terminal Tab (also Ctrl+Shift+] / Ctrl+Shift+[) |
Alt+Left / Alt+Right |
Switch Chat column tabs (Chat / Log / Stats) |
Alt+Shift+Left / Alt+Shift+Right |
Switch Quark Roster tabs (Active / All) |
Alt+PageUp / Alt+PageDown |
Switch Right-Rail Inspector tabs (Terminal / Files / Changes / Plan / Visualizer) |
Alt+Up / Alt+Down |
Cycle Telemetry & Stats time windows (Session / Week / Month / All) |
Ctrl+J / Ctrl+K |
Move selection in Quark Roster |
Ctrl+Alt+Enter |
Toggle selected Quark enabled / disabled |
Ctrl+O / Cmd+O |
Open Quick Project Switcher modal |
Ctrl+M / F10 |
Open App Menu overlay |
F6 |
Cycle global permission mode (Ask / Write / Auto / Bypass) |
📄 License & Community
- License: Apache License 2.0
- Contributing: CONTRIBUTING.md
- Security: SECURITY.md
- Code of Conduct: CODE_OF_CONDUCT.md
- Glossary: Physics Metaphor & Vocabulary
- Commands: Chat Commands Reference
Hadron MCP questions
How does Hadron ensure code safety?
Agents work in isolated git worktrees, and an automated Merge Gate rebases changes onto main and runs your test suite before code is merged.
Can I run Hadron in a headless environment?
Yes, Hadron provides standalone 'hadron run' and 'hadron ci' runners for headless execution in CI or terminal batch mode.