Claude Code mod vs MCP server vs skill vs hook: where Sume fits

Use an MCP server to give Claude Code Sume's tools, a skill to teach it how to use them, and a settings hook or a mod to guard or watch the calls. Claude Code's mods overview draws the same line: an MCP server is an external service that gives Claude tools and is the pick when Claude needs to reach an external system, while a mod is functions that run inside Claude Code and can change tool calls, prompts, and what the interface draws. Sume is an external system, so Sume itself arrives as an MCP server.

Mods are new (Claude Code v2.1.287 and later, on by default), which is why the question comes up now. Nothing about them replaces the hosted connector at https://mcp.sume.com/mcp.

How do the four compare?

Claude Code's overview has a four-way comparison of what each extension point is and when to pick it. The table keeps its columns and adds a last column for what each one means when the thing you want to reach is Sume.

The last row is the one that decides most questions. A mod, a hook and a skill all assume the capability already exists somewhere. Only the MCP server brings a new capability into the session, which is why a guard mod without a connected Sume server has nothing to guard.

Where does the hosted MCP fit?

It is the only one of the four that gives Claude new Sume capabilities. Sume's MCP overview lists what the hosted endpoint covers: account and catalog reads, jobs and assets, generate_image and generate_video, music_create, tts_create, stt_create, Avatar tools, and the crawl_* family. Two Sume products are REST-only and not in tools_list: Image 1.0 and Video 1.0, which you reach through the Developer API.

A plugin can hold several of these at once. Claude Code says a mod can ship in the same plugin as a skill and an MCP server, so a team can distribute the Sume connection and a guard mod together. The plugin `.mcp.json` post covers the connection half.

When is a skill better than a mod?

When the problem is knowledge rather than control. Sume's CLI documents sume skills install, which writes the packaged Sume skill into .agents/skills or .claude/skills, and sume skills export sume to review the files first. A skill changes what Claude knows and does; it cannot refuse a call. If you keep pasting the same Sume instructions into chat, that is a skill. The MCP versus agent skills post goes deeper.

When is a settings hook better than a mod?

When you already have a script. Claude Code's comparison says to pick a settings hook to block, allow, or log an event with a script you already have, and a mod when you want a pane, a band above the prompt, a custom command, or to rewrite an event. A PreToolUse shell hook that checks Sume's max_spend_usd field is a fine answer. Claude Code also notes that mods and settings hooks run in a fixed order, with managed-settings PreToolUse hooks first and a block from them final.

What should I pick for common jobs?

Here is the short version for the jobs developers ask about most. Each line names one mechanism and, where it helps, links to a worked example.

If you are unsure, start with the connection and add the others only when a specific problem shows up. Connecting takes two commands from the quickstart, and a first read-only call such as mcp_health or tools_list tells you the session works before any of the rest matters.

• Generate an image or clip from Claude Code: connect the hosted MCP.
• Make Claude follow your Sume prompt conventions: install a skill.
• Refuse any paid Sume call without max_spend_usd: a settings hook or a mod, depending on whether you already have a script.
• Show how many paid calls this session has made: a mod, because only a mod can draw. A counter mod is a worked example.
• Keep Sume credentials away from a mod: use OAuth, and read the credential post.