The definitive 2026 Claude Code reference: every slash command, MCP transport setup, hooks JSON schema, keyboard shortcuts, and recently-shipped features lik...
Slash Commands: The Fast Path Through Every Session
Slash commands are the quickest way to steer Claude Code without leaving the conversation. Type / and the picker lists what's available in the current directory, so you rarely need to memorize names. The core set covers the things you reach for constantly: clearing context when a thread gets long, switching models mid-task, reviewing a diff, checking permissions, and opening configuration. Because commands resolve against your working directory, a project can ship its own set alongside the built-ins.
Custom commands are just Markdown files whose body becomes the prompt. Drop a file in your project's commands folder and it shows up in the picker as /<name>; put it in your user-level folder and it follows you across every repo. This is the cleanest way to encode a repeatable workflow — a review checklist, a release routine, a scaffolding step — so it runs the same way every time instead of being re-explained.
MCP Transports: Connecting External Tools
The Model Context Protocol lets Claude Code talk to external servers — issue trackers, databases, browsers, internal APIs — through a standard interface. The main decision when adding one is the transport. Local servers usually run over stdio: Claude Code launches the process and speaks to it over standard input and output. Remote servers use HTTP-based transports, where you point at a URL and the server runs somewhere else, often behind auth.
- stdio — best for tools that live on your machine; you supply the command and arguments to launch.
- HTTP / streaming — best for hosted services; you supply a URL and any headers or tokens the server needs.
Register servers per project or at the user level depending on whether the connection is repo-specific. Scope matters for both convenience and safety: a database server tied to one project shouldn't be globally available. Once connected, an MCP server's tools appear to Claude the same way built-in tools do.
Hooks: Deterministic Control Over the Loop
Hooks run your own shell commands at defined points in a session — before or after a tool call, when a session starts, or when Claude stops. Unlike instructions in a prompt, hooks execute deterministically, which makes them the right tool for guarantees rather than suggestions: run a formatter after every edit, block writes to protected paths, or log activity. A hook that exits non-zero can stop an action, so hooks double as guardrails.
Hooks are configured as JSON in your settings, keyed by event and matched to specific tools. Each entry names the event, an optional matcher for which tool it applies to, and the command to run. Keep the commands fast and side-effect-aware — they run inline, so a slow hook slows every matching action.
Keyboard Shortcuts and Staying Efficient
The shortcuts that save the most time handle interruption and context. You can stop generation mid-stream to redirect before Claude finishes a wrong path, and edit or resubmit rather than starting over. Managing context length directly — clearing when a task is done — keeps responses sharp and costs down, since a bloated conversation makes every turn heavier.
Treat these features as one system rather than isolated tricks: a custom slash command can invoke tools exposed by an MCP server, and a hook can enforce policy on whatever that command does. Wiring commands, MCP, and hooks together is what turns Claude Code from a chat window into a repeatable part of your workflow.