Local-first MCP control plane for coding agents. Persistent memory across sessions so agents don't start blind. Token-aware file routing via knapsack optimizer. Orchestrator auto-selects from 90+ skills, 60+ workflows, and 7 crew templates across 30 domains. Phantom workers execute in isolated git worktrees. Governance blocks risky mutations. Runtime dashboard shows memory topology, token telemetry, and execution truth. Works with Claude Code, Cursor, Windsurf, Codex, Opencode. MIT licensed.

Hey Product Hunt — I built Nexus Prime because I kept hitting the same wall with AI coding tools: every new session started with re-explaining the codebase, rebuilding context, and wasting tokens just to get back to where I already was. Nexus Prime is an open-source MCP server built to reduce that tax. It gives AI coding agents: - persistent memory across sessions - token-aware context assembly - parallel execution using isolated Git worktrees - guardrails, workflows, and live observability - create and inject RAG during runtime The goal is simple: make coding agents feel less stateless and more durable. I would especially love feedback on three things: -Where does cross-session memory help most in your workflow? -Is token/context waste a real pain for you today? -What kinds of coding tasks would you trust parallel agents with? Would love your honest take

for discussion over memory, orhcestration, RAG, Git worktrees, tokens -
reddit https://www.reddit.com/r/Nexus_Prime/
discord https://discord.gg/zCPEjPZj

Nexus-Prime

Your coding agent memory, orchestration & runtime discipline

Your coding agent memory, orchestration & runtime discipline