Proof in the Open: What MPC Has Built and Who Found It
What has Mineral Project Coin (MPC) published as open source? This 13 November 2026 Building in Public essay lists the team's most-starred MIT-licensed tools for AI coding agents and BNB Chain developers, gives dated GitHub, npm and pub.dev figures, and names the projects they are built on.
There is a small moment we did not expect to feel so much. Someone we have never met opens a repository we wrote, decides it is worth keeping, and presses a star. No message, no name we recognise. Just a sign that the work reached someone and was useful to them.
We have had a few hundred of those moments now. This piece is about what we built to earn them, who we built it on, and why we think it matters for a mining real-world asset (RWA) project.
Why does a mining RWA team publish its tools in the open?
In July we wrote that the time for vibes was over and the time for proof had begun. We promised four things: time tracked, audits independent, evidence public, everything measured.
Promises are cheap. A team that asks to be trusted with records of real mines should be able to show its work before anyone has to take its word for it. So we made a rule for ourselves: when we solve a problem while building MPC, and the solution would help someone else, we publish it. Code, tests, and the numbers behind any claim we make about it.
Open source is the most honest place we know to keep that rule. The code is there. Anyone can read it, run it, and tell us where it is wrong.
What have we built in the open so far?
Everything below lives under github.com/globalmpc, under an MIT licence. We are a 100% AI-native team, so we met the failure modes of AI-written software before most people did. The tools people have picked up most are the ones we first wrote for ourselves, listed here by GitHub stars.
- nomore (104 stars). Finds the instructions you keep repeating to Claude Code and, with your approval, turns them into
CLAUDE.mdrules so they survive compaction and new sessions. It runs on your machine, uses no LLM, and ships a pack of 20 rules drawn from over 8,000 Claude Code GitHub issues. It also reads Codex CLI sessions. - vibe-preflight (65 stars). A pre-flight check for an AI-built app before it goes live: leaked keys, routes without login, packages the AI invented, paid APIs anyone can call, unsafe container and agent settings. It explains what it finds in plain language and changes nothing until you say so.
- backendskills (53 stars). Agent skills for backend code that is safe on auth, data, money and schema changes. In our benchmark of 15 tasks, runs with a critical mistake fell from 25 of 150 without the skills to 2 of 150 with them. The repository publishes the per-task table, the extra cost, and what is not fixed yet.
- jev-mem (22 stars). Brings back only the memory from past sessions that matters to the prompt you just typed, instead of loading tens of thousands of tokens every time.
- jev-router-effort (15 stars). Routes each coding-agent turn to the cheapest model and reasoning effort that is good enough.
- Just published: agent-leakcheck shows what a coding agent can reach on your machine, vibe-bill estimates what an AI-built app will cost to run, and open-agent-harness is a self-hosted team assistant that cites its sources.
For teams building on BNB Chain. rwa-kit-js (33 stars, seven npm packages), rwa-kit-flutter (23 stars, two pub.dev packages) and the public list of RWA engineering problems (21 stars) came straight out of building MPC. We have written about them in our dev updates: EVM Log Walker, Mnemonic Backup Flow and dapp-stats.
Our own product, in public. The MPC Registry (19 stars), the evidence registry at the centre of what MPC is building, is developed in the open with its build, coverage and security scans visible. It is in development: its contract runs on opBNB testnet and has not been deployed on BNB Smart Chain mainnet.
How have people responded?
We want to be careful here, because this is exactly where projects tend to round up.
As of 13 November 2026, our public repositories hold 381 GitHub stars in total. In the last 30 days nomore was downloaded 565 times from npm, the seven rwa-kit-js packages 2,822 times, and the two Flutter packages 721 times from pub.dev.
Those are not large numbers, and download counts include automated installs, so they are a ceiling on real use rather than a count of people. We will keep publishing them with a date next to them, the same way dapp-stats puts a block range next to every figure.
But we will admit what the numbers feel like from the inside. For a small team that has not shipped its main product yet, each star is a stranger telling us the work held up when they tried it. It is the first outside evidence that the way we build is worth something to people who owe us nothing.
Why do we name the work we build on?
Almost nothing we published started from zero. jev-mem is built on claude-mem and uses JEV as its judge. jev-router-effort is a derivative of jev-router and says so in its first lines. open-agent-harness is built on Nous Research's Hermes Agent. agent-leakcheck points readers to snyk/agent-scan and tells them to use both. nomore's rule pack exists because thousands of people took the time to file Claude Code issues. Underneath the chain packages are viem, OpenZeppelin and Foundry.
We name them in our READMEs because it is true, and because of what happens next. When someone finds one of our tools, they find the project it stands on. When someone goes looking for that project, some of them find ours, and through it, MPC. Light passes from hand to hand. Their work becomes more visible because we built on it; we become visible because they built first.
That is the kind of exposure we want for MPC. Not a campaign, but a trail of useful things that leads back to us, with every step checkable.
What does this have to do with mining RWA?
Everything, we think. The question at the centre of mining RWA is not whether an asset can be represented on chain. It is whether anyone can check what is said about it: who verified what, when, and whether the record kept up with the mine.
We are asking for the same kind of trust in our engineering that we will ask for in our records. So we hold the two to the same standard: write the claim down, attach the evidence, date it, and let anyone check it. Our open-source work is where you can watch us keep that standard today, while the platform itself is still being designed and built.
What comes next?
We will keep proving it. More of the problems on our public list will get packages. Every number we publish will carry its date and its source. When something we released turns out to be wrong, we will say so in the same place we announced it.
If one of these tools saved you an afternoon, a star helps the next team find it. If one of them broke, open an issue. Both are evidence, and we would rather have the evidence than the applause.
All repositories: github.com/globalmpc. Figures in this piece were read from GitHub, npm and pub.dev on 13 November 2026.