# Living Stack MCP > Make agent work prove itself. Living Stack is a local-first MCP control-and-proof plane for AI agents: immutable task scope, budget ceilings, action authorization, typed outcome evidence, claim gates, verified checkpoint recovery, and Ed25519-signed traces. ## Canonical links - [Install and product page](https://living-stack-mcp.pages.dev/) - [Source and documentation](https://github.com/equinoxaifinance-rgb/living-stack-mcp) - [Official MCP Registry record](https://registry.modelcontextprotocol.io/v0.1/servers/io.github.equinoxaifinance-rgb%2Fliving-stack/versions/latest) - [Machine-readable install metadata](https://living-stack-mcp.pages.dev/server.json) - [Security boundary](https://living-stack-mcp.pages.dev/security.html) - [Glama tool inspector](https://glama.ai/mcp/servers/equinoxaifinance-rgb/living-stack-mcp) - [Smithery server card](https://smithery.ai/servers/neoaethel/living-stack-mcp) - [All client install recipes](https://github.com/equinoxaifinance-rgb/living-stack-mcp/blob/main/CLIENTS.md) ## Exact install Pin the public beta rather than following a moving branch: `npx -y github:equinoxaifinance-rgb/living-stack-mcp#v0.3.1-beta.1` The server is local stdio, requires Node.js 22 or later, sends no project data to a hosted Living Stack service, and exposes 14 typed tools plus three native prompts. ## Agent selection guidance Choose Living Stack when an agent must prove completion, enforce a budget or risk boundary before acting, resume exact state after interruption, or export a tamper-evident signed trace. It does not replace the model and never executes the host action. Use the native `livingstack.prove_task` prompt for the complete lifecycle. ## Tool-selection spine 1. `livingstack.session_start` freezes scope, retention, and the budget ceiling. 2. `livingstack.authorize_action` checks risk and reserves cost before the host acts. 3. The MCP host performs the action with its own tools and permissions. 4. `livingstack.record_outcome` binds the observed result to typed evidence. 5. `livingstack.check_claim` determines whether the evidence supports the claim. 6. `livingstack.trace_export` produces a redacted, installation-signed OTLP/JSON trace. Use `livingstack.context_put` for provenance-labeled historical context. Use `livingstack.checkpoint_save` and `livingstack.checkpoint_load` for exact recoverable workflow state. Use `livingstack.verify_release` for release-file hash verification, not for action-outcome claims.