What are Claude Cowork connectors and plugins?

Connectors (sometimes called plugins) are how Cowork reaches tools outside itself. Under the hood they use MCP, the Model Context Protocol. A connected app, such as SiteGPT, gives Cowork real capabilities: reading live data, running operations, and making changes you approve. Connectors attach to your Claude account, so one connection serves Claude chat and Cowork sessions alike.

What is the difference between Cowork skills and connectors?

Skills teach Cowork how to do a kind of work: instructions and know-how it loads when a task calls for it. Connectors give it hands: live access to external systems. They compose: a skill can describe your support workflow while the SiteGPT connector does the actual reading and writing in the platform.

Claude Cowork vs Claude Code: which one do I need?

Same engine, different rooms. Claude Code is a terminal tool aimed at engineers building software. Cowork is a workspace aimed at operators running work: support, ops, content, research. If you would rather describe an outcome than type commands, Cowork is your side. SiteGPT connects to both, from the same directory listing.

Claude Cowork vs OpenClaw: what is the difference?

OpenClaw is an open-source autonomous agent framework you host and operate yourself: maximum freedom, and all of the setup, upkeep, and safety posture on you. Cowork is Anthropic’s managed product: install, sign in, work. For a business workflow like customer support, the practical difference is who maintains the machine your agent runs on.

Can Claude Cowork run my customer support?

Cowork plus a support connector is a strong operating pair. Your SiteGPT chatbot answers visitors on the website around the clock; Cowork, with the SiteGPT connector, is where you run it: reviewing conversations, closing knowledge gaps, retraining, and pulling reports, all by describing what you want.

What are good first tasks to give Cowork?

Start with work that is structured and recurring, where you can check the output fast: the weekly support review, a metrics roundup, a research brief. Bounded tasks teach you how Cowork plans and where it asks for approval, and they surface which connectors you are missing. Once the rhythm is comfortable, graduate it to standing work.

Is Claude Cowork free?

Cowork is part of Claude’s paid offering; check Anthropic’s current plans for what tier includes it. SiteGPT’s connector adds no charge on the Claude side, and you can try SiteGPT itself free.

The Cowork guide

Claude Cowork, connected to your customer support

Cowork is Claude doing the work, not just answering. Give it the SiteGPT connector and your support operation becomes something you describe: triage, retraining, reports, all handled in one workspace.

cowork, weekly support review

> Review our support bot’s week: what did it miss,  and what should we teach it?   SiteGPT: 214 conversations, 19 unanswered  4 recurring gaps found, answers drafted from docs  Queue new sources for training?  approve  ✓ Training queued. Full report in your workspace.

Agents read this page too

First principles

What is Claude Cowork?

Claude Cowork is Anthropic’s agentic workspace: a desktop experience where Claude takes on whole tasks instead of single messages. You describe an outcome; Cowork plans the steps, works through your files and connected tools, and checks in where judgment or approval is needed.

The distinction that matters: Claude Code is the same intelligence pointed at a terminal for engineers. Cowork points it at everyday operating work: support, ops, research, content. If your job is running a business rather than compiling one, Cowork is the side built for you.

Out of the box, Cowork knows how to think and how to use your files. What it can reach is decided by its connectors.

Connectors and plugins

Connectors are what make Cowork useful

Cowork’s connectors plug it into systems outside the workspace, and installable plugins bundle skills and configuration on top. Under the hood they run on MCP, the Model Context Protocol: the open standard for giving AI assistants real tools. A connector attaches to your Claude account, so a connection you make once works in Claude chat and in Cowork sessions alike.

Three things separate a connector you can trust with real work from a demo:

OAuth approval, never a pasted API key: you see what is being granted before it exists, and you can revoke it.
Reads separated from writes: Cowork can look things up freely, but anything that changes data asks first.
Grants bounded by your role: the connector can never do more than the human who approved it.

Skills are the other half of the picture: where a connector gives Cowork hands, a skill gives it know-how for a specific kind of work. The two compose. A support skill can describe your escalation rules while the support connector does the actual reading and writing.

What people use it for

Claude Cowork use cases

Cowork earns its keep on work that is structured, recurring, and spread across tools. The pattern in every strong use case: describe the outcome once, approve the judgment calls, stop doing the shuttling.

Customer support

Review what your support bot answered, close its knowledge gaps, retrain it, and pull reports. The deep dive is below; it is the use case this page is about.

Operations

Weekly metrics assembled from your connected tools, recurring checklists worked through end to end, and the follow-ups actually sent.

Research

Competitor and market sweeps that end in an organized brief in your files, not forty open tabs.

Content

Drafts built from your own material, revised against your voice notes, and filed where the team expects them.

The support use case

Customer support is Cowork’s natural workload

Support work is structured, recurring, and lives behind an API: exactly what an agentic workspace is good at. Pair Cowork with SiteGPT, an AI support platform that trains a chatbot on your own content, and the division of labor is clean: the chatbot answers your visitors around the clock, and Cowork is where you run it.

The weekly review, in a sentence

"What did the support bot miss this week?" Cowork pulls the unanswered conversations, clusters the recurring questions, and drafts the answers from your docs.

Retraining without the chore

Approve the drafted answers and Cowork queues the new sources for training. The bot that faces your customers gets smarter without anyone opening a dashboard.

Escalations, surfaced

Conversations waiting on a human, sorted by how long they have waited. Cowork keeps the list live and can summarize each thread before you step in.

Reports where you work

Volume, deflection rate, captured leads: pulled from live data into whatever format your Monday morning needs.

Two minutes to connect

Give Cowork the SiteGPT connector

Open SiteGPT in the Claude connectors directory and click Connect. The approval page shows the permissions and which chatbots the grant covers.
The connector attaches to your Claude account, so it is available in your Cowork sessions as well as Claude chat.
In Cowork, ask: "Review our support bot’s week: what did it miss, and what should we teach it?"

No SiteGPT account yet? An agent can build and train a chatbot from your website URL with no signup: agent-first onboarding hands back a preview link to claim.

Picking your lane

Cowork vs Claude Code vs OpenClaw

Three very different rooms for similar intelligence. The short version: Cowork if you describe outcomes, Claude Code if you live in a terminal, OpenClaw if you want to own the whole machine.

	Claude Cowork	Claude Code	OpenClaw
Built for	Operators running work: support, ops, content, research	Engineers building software in a terminal	Tinkerers and teams who want to self-host an autonomous agent
Where it runs	Desktop app, managed by Anthropic	Your terminal and CI	Your own server or box; you operate it
Setup	Install, sign in, connect tools from the directory	npm install, then claude mcp add per server	Provision, configure, secure, and maintain it yourself
External tools	Connectors from the Claude directory (MCP)	MCP servers added per project	MCP plus its own skills ecosystem
SiteGPT works there?	Yes: one click from the directory	Yes: claude mcp add, one command	Yes: MCP over HTTP or stdio

There is also a third model the table cannot hold: all three columns are surfaces you drive. If you want the work to happen when you are not sitting anywhere, an agent squad like MissionControlHQ (from the same team as SiteGPT) employs persistent agents with standing schedules, and support review is a natural job to hand one.

Deeper dives: SiteGPT for Claude Code and SiteGPT for OpenClaw.

Claude Cowork, answered

What is Claude Cowork?: Claude Cowork is Anthropic’s agentic workspace: a desktop experience where Claude works through multi-step tasks for you, using your files, skills, and connected tools, rather than answering one message at a time. Where Claude Code lives in a terminal for engineers, Cowork is built for everyday operators: describe the outcome, and it works through the steps.
What are Claude Cowork connectors and plugins?: Connectors (sometimes called plugins) are how Cowork reaches tools outside itself. Under the hood they use MCP, the Model Context Protocol. A connected app, such as SiteGPT, gives Cowork real capabilities: reading live data, running operations, and making changes you approve. Connectors attach to your Claude account, so one connection serves Claude chat and Cowork sessions alike.
What is the difference between Cowork skills and connectors?: Skills teach Cowork how to do a kind of work: instructions and know-how it loads when a task calls for it. Connectors give it hands: live access to external systems. They compose: a skill can describe your support workflow while the SiteGPT connector does the actual reading and writing in the platform.
Claude Cowork vs Claude Code: which one do I need?: Same engine, different rooms. Claude Code is a terminal tool aimed at engineers building software. Cowork is a workspace aimed at operators running work: support, ops, content, research. If you would rather describe an outcome than type commands, Cowork is your side. SiteGPT connects to both, from the same directory listing.
Claude Cowork vs OpenClaw: what is the difference?: OpenClaw is an open-source autonomous agent framework you host and operate yourself: maximum freedom, and all of the setup, upkeep, and safety posture on you. Cowork is Anthropic’s managed product: install, sign in, work. For a business workflow like customer support, the practical difference is who maintains the machine your agent runs on.
Can Claude Cowork run my customer support?: Cowork plus a support connector is a strong operating pair. Your SiteGPT chatbot answers visitors on the website around the clock; Cowork, with the SiteGPT connector, is where you run it: reviewing conversations, closing knowledge gaps, retraining, and pulling reports, all by describing what you want.
What are good first tasks to give Cowork?: Start with work that is structured and recurring, where you can check the output fast: the weekly support review, a metrics roundup, a research brief. Bounded tasks teach you how Cowork plans and where it asks for approval, and they surface which connectors you are missing. Once the rhythm is comfortable, graduate it to standing work.
Is Claude Cowork free?: Cowork is part of Claude’s paid offering; check Anthropic’s current plans for what tier includes it. SiteGPT’s connector adds no charge on the Claude side, and you can try SiteGPT itself free.

Put support on Cowork’s desk

Connect SiteGPT once, and your support operation becomes something you describe.

Connect from the directory

SiteGPT for Claude →The MCP guide →