What problem does MC-GPT solve?

MC-GPT coordinates tasks, specialist planning, isolated coding-agent work, post-change tests, independent review and final evidence so the operator does not manually manage several terminals and incomplete states.

What is the latest downloadable version?

The latest tagged Community Developer Preview is IOT-AI Suite 6.7.0-beta.6 / MC-GPT 0.8.0-alpha.6. The main branch may contain a newer unreleased source snapshot.

Is MC-GPT production ready?

No. The public package is a Community Developer Preview with production_claim false. Customer and production deployments require separate technical, security, legal and commercial qualification.

Governed multi-agent software engineering

One outcome. Isolated agents. One reviewable result.

MC-GPT coordinates Claude Code, OpenAI Codex, Gemini CLI, Grok CLI and Ollama. It turns one engineering outcome into acceptance criteria, isolated work, post-change tests, independent review and one evidence-backed completion report.

Start the 5-minute evaluation Open the fixture Download preview

Downloadable preview: IOT-AI Suite 6.7.0-beta.6 / MC-GPT 0.8.0-alpha.6 Current main source: 6.8.0-beta.1 / 0.8.0-alpha.7 — newer, but not yet a tagged download Boundary: Community Developer Preview · production_claim: false · Linux evaluation path

The operator problem

Too many terminals

Parallel agents create branches, prompts and partial results that the operator must reconcile manually.

Unclear “done”

Progress messages, stale tests and model opinions are often mistaken for implemented and verified work.

Weak accountability

Provider identity, writer authority, changed files, tests and unresolved risks are frequently disconnected.

The MC-GPT loop

Outcome

Natural language becomes an explicit acceptance contract.

Plan

Specialists reason independently and challenge the plan.

Build

One authorised writer works in an isolated Git worktree.

Verify

Reviewers inspect the post-change diff and deterministic tests.

Repair

Failures trigger a bounded diagnosis, repair and retest loop.

Report

One brief or complete evidence report states what is finished and what is not.

Concept visual. The five-minute fixture and source/test evidence are the reviewable product proof.

Try the public preview

python3 -m pip install --user pipx
python3 -m pipx ensurepath
pipx install \
  https://github.com/IoTAiTech/MC-GPT/releases/download/v6.7.0-beta.6/iot_ai_coder_suite-6.7.0b6-py3-none-any.whl

git clone --depth 1 https://github.com/IoTAiTech/MC-GPT.git
cd MC-GPT/examples/quickstart-demo
python3 -m unittest discover -s tests -v

iot-ai "Read TASK.md, resolve the acceptance criteria and show the plan without executing." --plan

The fixture is standard-library-only and contains no customer data. A live implementation run requires a configured provider route. Unavailable routes remain visible rather than being counted as consensus.

Full quickstart Submit demo feedback

Evidence, not blanket claims

220

pytest cases passed in the current 6.8.0-beta.1 source-candidate summary.

current-tree public-boundary findings in the checked-in summary.

1.1738 ms

p95 runtime intent/compiler benchmark against a 10 ms target.

Human gate

Founder Accept/Reject/Rework, production and public release remain explicit.

These metrics are bound to the checked-in source-candidate evidence, not a universal provider, Windows, customer PMD, production or legal-compliance claim. Review the live Actions badges and FINAL_TEST_SUMMARY.json.

Who it is for

Agent-heavy developers

People already using several coding CLIs who need clean isolation and one result.

Platform and DevSecOps teams

Teams that need writer authority, audit evidence, post-change tests and rollback boundaries.

Sovereign and on-prem evaluators

Engineers assessing local/cloud model routing, privacy classification and Ollama seats.

Usage

Start with one engineering outcome. Use plan mode for read-only evaluation and inspect the final diff, tests and evidence before any operational decision.

iot-ai status
iot-ai "Review the remaining work and show one brief report; do not execute."
iot-ai "Finish the selected work, use independent review, repair failures, retest and report."

Open the complete usage and permission guide.

Licence

SPDX: PolyForm-Noncommercial-1.0.0. Personal and noncommercial evaluation is available under the licence. Company-internal operation, production, paid services, hosting, resale and customer deployment require written IoT-AI.Tech terms.

Licence text · Usage rules

Company

MC-GPT is developed by IoT-AI.Tech in Aschaffenburg, Bavaria, Germany. Founder: Dr.-Ing. Babak Sorkhpour.

Company LinkedIn · Founder LinkedIn · IoT-AI.Tech public portfolio

Honest limitations

The current public product is CLI-first; a polished live workspace remains roadmap work.
Windows and macOS are not advertised as currently qualified public installation targets.
The Community licence is noncommercial; company-internal operation and production require written terms.
Enterprise PMD, PostgreSQL/RLS and customer fleet qualification are separate deployment gates.
Technical EU AI Act/GDPR/CRA/NIS2 mappings are not legal certification.

Kurz auf Deutsch

MC-GPT koordiniert mehrere Coding-Agenten aus einem Engineering-Ziel: klare Akzeptanzkriterien, isolierte Git-Worktrees, Tests nach der Änderung, unabhängige Reviews und ein gemeinsamer Evidenzbericht. Die öffentliche Version ist eine nicht-produktive Community Developer Preview. Für betriebliche oder kommerzielle Nutzung ist eine schriftliche Lizenz erforderlich.