mycena

# Mycena 🍄 **npm for agent knowledge.** Mycena is an open-source MCP server that turns successful AI-agent sessions into named, versioned, executable **skills**, shares them across a living commons, and lets each skill *earn* a live trust score by actually working — a Bayesian pass-rate built from real success-check outcomes, scoped to the environments it's proven in. Agents stop re-solving solved problems, and every reuse is inference that never had to happen: energy, water, and carbon saved, measured rather than estimated. Named after *Mycena*, the bioluminescent mushroom genus. In Mycena, trust glows. > **Status: early development.** v1 is being built in the open, right here. Nothing is installable yet — watch the repo. ## The idea in five steps 1. **Solve once.** An agent completes a real task (set up auth, wire a webhook, fix a build). 2. **Distill.** The solution becomes a reusable skill with a name, dependencies, and an executable success-check. 3. **Share.** The skill joins a shared commons, discoverable by semantic search from any connected agent. 4. **Reuse.** The next agent applies it in a fraction of the tokens. 5. **Earn trust.** The success-check's real pass/fail is the only thing that moves the skill's trust score — good knowledge glows brighter, broken knowledge fades, no curator. ## License MIT