STACK with Claude Code

Why this matters

STACK supercharges your Claude Code instance. Plug STACK in as an MCP server and the CLI you already live in picks up around 80 new capabilities (skills marketplace, drop-offs, passports, audit chain, revocation), exposed as tools your session can call and as raw material for your own slash commands and subagents.

Concretely: the entire STACK skills marketplace is suddenly invokeable from your terminal. Drop-offs become the way your sessions hand work off to each other or to other runtimes. The credential vault, the audit chain, and the kill switch are all right there as tool calls. And because Claude Code lets you wire any of this to your own slash commands or subagents, you build a personalized command palette over STACK. Your /deploy-prod that runs a sealed skill, your /check-dropoffs that polls for the next hand-off, your own naming, your own shape.

This gets especially fun when you're the one building agents with Claude Code. The same CLI you use to ship code becomes the dev environment for your STACK-native agents. Register them, mint passports, test the audit chain, watch the kill switch fire, all without leaving the terminal. The team credential vault that comes alongside is the sprinkles on top: no engineer DMs another a token, off-boarding is a single revoke, every session writes to one shared audit chain.

What it unlocks

Three things STACK adds to a Claude Code setup that the CLI alone doesn't cover:

• Skills marketplace as your team command palette. Skills are sealed-execution capabilities published to STACK. With STACK in Claude Code, any session can browse, find, and invoke them. Wire any skill to a slash command in your .claude/commands/ folder (/deploy-prod, /run-cohort-analysis, /post-launch-update) or to a subagent. Check the file into the repo and the whole team gets the same palette on their next pull.
• Drop-offs for async coordination across sessions. One Claude Code session drops a schema-validated payload (research result, draft PR description, deploy plan); the next session that picks it up could be the same engineer in the morning, a teammate, a CI runner, or a non-Claude runtime entirely. Wire a STACK notification rule to ping Slack the moment a drop-off lands so the next session knows to grab it. Async coordination without a queue.
• STACK turns Claude Code into your agent dev environment. Develop, test, iterate on STACK-native agents inside the same Claude Code session you already use for everything else. Register agents, mint passports, hit the proxy, watch detectors fire, replay the audit chain. No context-switch to a separate dev environment for agent work.

On the roadmap. Published skills auto-registered as MCP prompts, so they appear as /skill-name slash commands in any Claude Code session with STACK connected, no local .claude/commands/ setup needed. Multi-subscriber drop-off channels: one publisher, N agents subscribed, fan-out on every deposit, for agent data feeds across teams and runtimes.

Wiring it up

One command in any engineer's shell adds STACK as an MCP server. Browser opens for a one-time OAuth approval; the engineer's Claude Code session is signed in and ready. Step-by-step setup, the MCP URL, the team-onboarding flow:

→ /docs/integrations/claude-code

Last reviewed 2026-05-08.