MCS puts multiple AI models, cloud and local, into one structured deliberation. They propose, challenge each other, defend their reasoning, and converge on an answer that survived scrutiny.
Multi-Model AI Reasoning, Deliberation & Decision Engine. Runs on your Linux or macOS machine.
MCS Engine is software that makes several AI models work on one problem together. Instead of trusting a single model's answer, MCS runs a structured deliberation: the models propose, challenge each other, defend their reasoning, and converge on an answer that survived scrutiny. And they collaborate with more than words: real files, such as images, PDFs, and code, pass between the models during the session. That is what MCS is for: decisions where being wrong is expensive. Code reviews, legal analysis, medical second opinions, research, hard engineering choices.
It installs on your own Linux or macOS machine, and you bring your own AI models, in any mix: all cloud, all local, or both at the same table. Cloud models need nothing but their API keys. Local models run through whatever you already use, Ollama, LM Studio, or any OpenAI compatible server. There is nothing else to install: MCS is one self contained download. MCS is not a model and does not host models. It is the engine that turns the models you already have into a team.
Other tools collect independent answers and pick one. That is a survey. In MCS, participants see each other's reasoning across multiple rounds and must defend or revise their positions. That is a meeting.
MCS runs on-premise. Mix cloud providers with local models running on your own hardware. Your data goes only where you send it.
Every session produces a full record of who proposed what, who challenged it, and why the conclusion held. Not a verdict, but a defensible position.
Code review, legal analysis, research, creative direction. Wherever a single opinion is not enough.
This is MCS running a legal strategy deliberation: a wrongful termination case, with the case brief attached as a reference file. Three models from three providers: Claude moderating, GPT-4o and Grok arguing opposing positions. No pipelines to build, no code to write, nothing to compile: you configure the panel in your browser and press start. These are unedited screenshots. Click any of them to enlarge. Every package ships with nine sessions like this one, ready to run.
One command in a terminal. MCS starts its web server and opens the interface in your browser.
Claude as moderator, GPT-4o and Grok with opposing roles, each with its own instructions, and the case brief attached as a reference file.
Live, turn by turn: the models argue the case, cite the brief, and challenge each other's numbers.
Full transcript, per-turn files, summaries, and the memo the panel produced, all as ordinary files in the session folder.
A client-ready strategy memo, produced as a real file. This session: nine turns, two rounds, 3 minutes 10 seconds.
OpenAI, Anthropic, Google, and local models through Ollama, LM Studio, or any OpenAI compatible server, all in the same deliberation, at the same table. Run private models for sensitive work, add cloud power when you need it. No vendor lock-in, ever.
MCS participants collaborate on images, documents, audio, source code, and data. Real files, passed between models during the discussion, so every decision gets richer context.
The SDK is a C API and a Python package that let your own programs start an MCS deliberation, watch its progress, and collect the results. It drives your installed MCS Engine, so sessions behave exactly as they do from the web interface. The SDK is included with Pro Max and Enterprise, and available as a $15/mo add on for Basic and Pro.
import time
import mcs
mcs.init() # finds your engine and license on its own
with mcs.start("mcs.conf", "session.conf") as session:
while not session.is_done():
time.sleep(2)
print("status:", session.status())
for name in session.output_files():
print("produced:", name)