One Plugin per Service

Anthropic's claude-tag-plugins repo doesn't hook Claude up to everything. Each service is its own plugin, so a workspace can connect exactly the services it uses — the real question isn't what's available, it's what your team actually works in.

Three terms first. A plugin: one service's skill and scripts, packed to install. An agent scope: Claude's settings for one workspace, or one channel. An identity profile: a reusable bundle of rules, credentials, and plugins.

Picture your team's workspace as a wall of sockets, one for each service it could use. Claude Tag is the Claude that answers when you tag it in a chat thread, backed by remote Claude Code sessions. The repo lists eighteen plugins — sixteen are services, from Asana and BigQuery to Jira and Datadog, and two are helpers. Add the repo as a marketplace, then install only the services you use. Say a team works in Asana and keeps its data in BigQuery — two plugs go in, every other socket stays empty. Tag Claude in a thread, and it can reach exactly those two: Asana tasks and SQL in BigQuery. Ask about Jira, and there's no Jira plugin to reach for.

Open one plug and there's a manifest, plugin.json; a skill describing how to connect and the core operations; a catalog of endpoints, read only when needed; and usually small scripts. What you won't find is a key — the runtime injects credentials into each request. Claude's settings come in layers: organization, then workspace, then channel, and plugins add up along that chain — a channel can plug in more, but can't unplug what a layer above provides, so what's plugged in at the top reaches every channel.

For GitHub, the repo's best practice goes further: a read-only profile bound everywhere, and a write profile bound only to the channels allowed to push or merge — limiting the blast radius. Plug in what you use, and give each plug the least it needs.

The two helpers: hand the data visualization plugin a table — say, rows from BigQuery — and it composes a polished chart as a PNG, SVG, or interactive HTML page. Troubleshoot is the other: config changes only reach new sessions, so start a fresh thread, then run debug-plugins from inside the session to check which plugins actually loaded, which failed, and why.

Your Turn: Paste into Claude: "Here are the tools my team uses each week: [list them]. From the plugin list in anthropics/claude-tag-plugins, pick the ones our Claude Tag needs. For each, say whether it should only read or also write, and in which channels. Then list what to leave unplugged, and why." Then check two things yourself: for every plugin, who actually used that tool this week, and for every write, which channel and task actually need it?

Source: Anthropic, claude-tag-plugins (README and plugin folders)

0:00 The question, and three terms 0:23 The workspace wall 0:34 Eighteen plugins 0:46 Plug in what you use 0:58 What Claude can reach 1:10 Inside a plug 1:24 Layers of scope 1:37 Read everywhere, write where needed 1:51 A table becomes a chart 2:03 Troubleshoot 2:15 Your turn

Part of the Claude & Agentic AI series. @NikBearBrown

Deliberately Not Claimed

This video summarizes Anthropic's own published README and plugin structure for claude-tag-plugins and does not claim to introduce new mechanics or supersede that source. The specific plugin count (eighteen) and the Asana/BigQuery example reflect the repository at the time of recording and may change as plugins are added or removed. Check the linked repository for the current, authoritative list and configuration details.

SEO Keyword Tags

claude-tag-plugins, Claude Tag, Claude plugins Slack, least privilege AI agent, Claude Tag Asana, Claude Tag BigQuery, agent scope workspace channel, read-only write profile GitHub, Claude Tag troubleshoot, Claude data visualization plugin, Claude plugin marketplace, Claude Tag setup, agentic AI least privilege, Claude Code remote sessions, Slack Claude integration

What this video covers

• 0:00 — The question, and three terms 0:23 The workspace wall 0:34 Eighteen plugins 0:46 Plug in what you use 0:58 What Claude can reach 1:10 Inside a plug 1:24 Layers of scope 1:37 Read everywhere, write where needed 1:51 A table becomes a chart 2:03 Troubleshoot 2:15 Your turn

Transcript

A lightly formatted transcript of the video, with timestamps so you can jump to any section on YouTube.

The question, and three terms 0:23 The workspace wall 0:34 Eighteen plugins 0:46 Plug in what you use 0:58 What Claude can reach 1:10 Inside a plug 1:24 Layers of scope 1:37 Read everywhere, write where needed 1:51 A table becomes a chart 2:03 Troubleshoot 2:15 Your turn

[0:00] Ola, this is Liam in for Bear. Anthropic's Claude tag plugins repo doesn't hook Claude up to everything. Each service is its own plugin, so the real question is what your team uses. A plugin one service is skill and scripts packed to install, an agent scope Claude settings for one workspace or one channel, and an identity profile, a reusable bundle of rules, credentials, and plugins. Here's your team's workspace drawn as a wall of sockets, one for each service it could use. And here's Claude Tag, the Claude that answers when you tag it in a chat thread backed by remote Claude code sessions. The repo lists 18 plugins. 16 are services from Asana and BigQuery to Jira and DataDog. The other two are helpers. You add the repo as a marketplace, then install the services you use. Each service is its own plugin, so a workspace can connect exactly the services it uses. Say this team works in Asana and keeps

Its data in BigQuery. Two plugs go in, every other socket stays empty. Now tag Claude in a thread, it can reach those two, Asana tasks and SQL in BigQuery. Each skill activates when it's relevant. Ask about Jira and there's no Jira plugin to reach for. Open one plug, there's a manifest plugin.json, a skill, how to connect and the core operations, a catalog of endpoints read only when needed, and usually small scripts. What you won't find is a key. The runtime injects the credentials into each request. Claude settings come in layers, the organization, a workspace, then a channel. Plugins add up along that chain. A channel can plugin more but can't unplug what a layer above provides. So, what you plugin at the top reaches every channel. For GitHub, the repo's best practice goes further, a read-only profile bound everywhere and a write profile bound only to the channels allowed to push or merge. That limits the blast radius. Plugin

What you use and give each plug the least it needs. Now the two helpers. Hand the data visualization plugin a table, say rows from BigQuery, and it composes a polished chart, a PNG, an SVG, or an interactive HTML page. The other helper is troubleshoot. Config changes only reach new sessions, so start a fresh Slack thread, then run debug plugins. From inside the session, it checks which plugins actually loaded, which failed, and why. Your turn. Paste this into Claude. Here are the tools my team uses each week. List them. From the plugin list in Anthropic's Claude tag plugins repo, pick the ones our Claude tag needs. For each, say whether it should only read or also write, and in which channels. Then list what to leave unplugged and why. Then check two things yourself. For every plugin, who used that tool this week? And for every write, which channel and task need it? One plugin per service at Nick Bear Brown.