One MCP server that gives every AI agent your identity, shared memory, and action guardrails. Install once, every agent gets smarter.
Identity, memory, and guardrails — the three things every agent needs but none of them have.
Your preferences, context, and rules — queryable by any agent. Define your coding style, communication tone, schedule constraints, and project context once.
Shared knowledge layer across all your agents. Agent A learns something, Agent B knows it. Persistent context that compounds over time.
Action rules and guardrails. Before any agent takes a consequential action, it checks Guard. Set spend limits, require approvals, block risky actions.
Define who you are, what agents should remember, and what they're not allowed to do.
identity:role: "Founder building AI infra startup"communication:team: casual, direct, emoji-friendlyclients: professional, conciseinvestors: data-driven, confidentcoding:lang: typescriptstyle: functionaltests: vitest, integration-firstschedule:no_meetings_before: "10:00"focus: ["Tue 9-12", "Thu 9-12"]memory:store: local # sqlite, local-firstretention: 90d # auto-expire old entriesnamespaces: [work, personal, side-projects]guard:rules:- spend_money:max_auto: $50notify_above: $20- send_external:require_approval: true- modify_production:blocked: true
One command. Creates memvex.yaml and starts the MCP server locally.
Edit memvex.yaml with your preferences, rules, and context. Or use the web UI.
Add Memvex as an MCP server in Claude, Cursor, or any MCP-compatible tool.
Every agent now queries your identity, shares memory, and respects your guardrails.
And anything that can make an HTTP call.
Open source. Local-first. MCP-native. Always free.