Salvor MCP alpha: Claude and ChatGPT sharing one project brain
Dan from Salvor Labs published the first full demo of Salvor MCP alpha today. Two AI tools from two different companies work on the same real project, and what one learns, the other can find, check and build on. This is the demo, and what to watch for.
The video
The same video is on the Salvor Labs YouTube channel: Salvor MCP Alpha Demo. A version with Turkish subtitles is on the Turkish page.
What happens in it
- A real $SALVOR payment on Solana mainnet is verified and turned into MCP service credits.
- A metered MCP call is rejected while the account has no credits, so the metering is real.
- Claude Code investigates a small app (TaskFlow), runs tests and saves an evidence-backed lesson, a Domain Learning, into the project.
- ChatGPT in the browser finds Claude's lesson through Salvor MCP, applies it to a new feature and contributes code, tests and its own lesson through GitHub.
- Claude checks the work on its own and merges it.
- Both lessons stay in the repository with their author, their evidence and their review status.
Why it matters
Chat memory helps an AI remember what happened. Salvor keeps what was learned, the evidence behind it, and enough context for the next agent, or the next person, to build on it. That knowledge belongs to the project, not to one vendor's model. Dan's short version: knowledge over memory.
Dan also notes that this is footage of real runs, edited for length.
Links
- Salvor Labs: salvorlabs.org
- Community: t.me/SalvorLabs
- Demo repository: SalvorLabs/salvor-taskflow-demo-app
Get new articles by email
Only when a new article is out. No ads. Details →