Launching today

SDD Vault
Specs your AI agents follow, and can't rewrite
1 follower
Specs your AI agents follow, and can't rewrite
1 follower
Store your constitution, specs, plans and tasks in one versioned place and serve them to every AI agent through MCP. Agents claim tasks so they don't collide, every write names the agent, and nobody can rewrite the rules. Free for 3 people.



Free Options
Launch Team

AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted
Hi Product Hunt, I'm Eduin, maker of SDD Vault.
I build software with Spec Driven Development: a constitution, then a spec, a plan and tasks, and AI agents implement the tasks. It works until you run more than one agent. Two agents on two branches start the same task. An agent asked to make a test pass edits the spec instead of the code. "Approved" is a word in a markdown header that nothing enforces.
SDD Vault is a server for those documents. Your agents read and write them through MCP instead of the file system:
An agent claims a task before it starts; the others see it taken and pick the next one. The app shows who is on what, live.
Every write is a revision with the person, the agent and the time. Nothing is overwritten.
The constitution is protected: only owners write it, and a refusal comes back as an error the agent can read.
"Which requirements have no task?" is one call, answered from an index, not by parsing files.
Import an existing SDD, GitHub Spec Kit or Kiro project in one command; export back to markdown, byte for byte.
We built SDD Vault with SDD Vault. Three weeks ago a Claude Code session moved our whole backend from Rust to TypeScript from a 62-task spec it could not edit: 1,141 of 1,142 tests ported, test suite from 46 minutes to 7. The write-up is here: https://dev.to/eduin_garciacorde...
Works with Claude Code, Cursor, Windsurf, VS Code, JetBrains and any MCP client. Free for 3 people and 1 project, no card.
What I would like to know from you: when you run agents in parallel, what breaks first for you?